docs

enable Edit on GitHub and last updated on all doc pages
rewrite docs root page, add screenshot
2026-09-27 06:05:14 -07:00 · 2026-03-16 05:11:21 -07:00 · 2026-03-16 04:45:40 -07:00 · 2026-03-16 04:12:11 -07:00 · 2026-03-16 04:09:50 -07:00 · 2026-03-16 04:08:46 -07:00
215 changed files with 9915 additions and 13379 deletions
@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.2.1
+current_version = 0.2.3
 commit = True
 tag = True
 tag_name = v{new_version}
@@ -61,6 +61,7 @@ jobs:
          python -m pip install --upgrade pip
          pip install pyinstaller
          pip install -r backend/requirements.txt
+          pip install --no-deps chatterbox-tts

      - name: Install MLX dependencies (Apple Silicon only)
        if: matrix.backend == 'mlx'
@@ -122,7 +123,7 @@ jobs:
          p12-file-base64: ${{ secrets.APPLE_CERTIFICATE }}
          p12-password: ${{ secrets.APPLE_CERTIFICATE_PASSWORD }}

-      - uses: tauri-apps/tauri-action@v0
+      - uses: tauri-apps/tauri-action@v0.6
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          TAURI_SIGNING_PRIVATE_KEY: ${{ secrets.TAURI_SIGNING_PRIVATE_KEY }}
@@ -173,6 +174,7 @@ jobs:
          python -m pip install --upgrade pip
          pip install pyinstaller
          pip install -r backend/requirements.txt
+          pip install --no-deps chatterbox-tts

      - name: Install PyTorch with CUDA 12.1
        run: |
@@ -49,6 +49,7 @@ logs/
 # Generated files
 app/openapi.json
 tauri/src-tauri/binaries/*
+tauri/src-tauri/gen/Assets.car

 # Temporary
 tmp/
@@ -66,14 +66,16 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - OpenAPI client generator script now documents the local backend port and avoids an unused loop variable warning

 ### Added
- **Makefile** - Comprehensive development workflow automation with commands for setup, development, building, testing, and code quality checks
-  - Includes Python version detection and compatibility warnings
-  - Self-documenting help system with `make help`
-  - Colored output for better readability
-  - Supports parallel development server execution
+- **justfile** - Comprehensive development workflow automation with commands for setup, development, building, testing, and code quality checks
+  - Cross-platform support (macOS, Linux, Windows)
+  - Python version detection and compatibility warnings
+  - Self-documenting help system with `just --list`

 ### Changed
- **README** - Added Makefile reference and updated Quick Start with Makefile-based setup instructions alongside manual setup
+- **README** - Updated Quick Start with justfile-based setup instructions
+
+### Removed
+- **Makefile** - Replaced by justfile (cross-platform, simpler syntax)

 ---

@@ -1,250 +0,0 @@
-# Voicebox Makefile
-# Unix-only (macOS/Linux). Windows users should use WSL.
-
-SHELL := /bin/bash
-.DEFAULT_GOAL := help
-
-# Directories
-BACKEND_DIR := backend
-TAURI_DIR := tauri
-WEB_DIR := web
-APP_DIR := app
-
-# Python (prefer 3.12, fallback to 3.13, then python3)
-PYTHON := $(shell command -v python3.12 2>/dev/null || command -v python3.13 2>/dev/null || echo python3)
-VENV := $(CURDIR)/$(BACKEND_DIR)/venv
-VENV_BIN := $(VENV)/bin
-PIP := $(VENV_BIN)/pip
-PYTHON_VENV := $(VENV_BIN)/python
-
-# Colors for output
-BLUE := \033[0;34m
-GREEN := \033[0;32m
-YELLOW := \033[0;33m
-NC := \033[0m # No Color
-
-.PHONY: help
-help: ## Show this help message
-	@echo -e "$(BLUE)Voicebox$(NC) - Development Commands"
-	@echo ""
-	@grep -E '^[a-zA-Z_-]+:.*?## .*$$' $(MAKEFILE_LIST) | sort | \
-		awk 'BEGIN {FS = ":.*?## "}; {printf "  $(GREEN)%-20s$(NC) %s\n", $$1, $$2}'
-
-# =============================================================================
-# SETUP
-# =============================================================================
-
-.PHONY: setup setup-js setup-python setup-rust
-
-setup: setup-js setup-python ## Full project setup (all dependencies)
-	@echo -e "$(GREEN)✓ Setup complete!$(NC)"
-	@echo -e "  Run $(YELLOW)make dev$(NC) to start development servers"
-
-setup-js: ## Install JavaScript dependencies (bun)
-	@echo -e "$(BLUE)Installing JavaScript dependencies...$(NC)"
-	bun install
-
-setup-python: $(VENV)/bin/activate ## Set up Python virtual environment and dependencies
-	@echo -e "$(BLUE)Installing Python dependencies...$(NC)"
-	$(PIP) install --upgrade pip
-	$(PIP) install -r $(BACKEND_DIR)/requirements.txt
-	$(PIP) install --no-deps chatterbox-tts
-	@if [ "$$(uname -m)" = "arm64" ] && [ "$$(uname)" = "Darwin" ]; then \
-		echo -e "$(BLUE)Detected Apple Silicon - installing MLX dependencies...$(NC)"; \
-		$(PIP) install -r $(BACKEND_DIR)/requirements-mlx.txt; \
-		echo -e "$(GREEN)✓ MLX backend enabled (native Metal acceleration)$(NC)"; \
-	fi
-	$(PIP) install git+https://github.com/QwenLM/Qwen3-TTS.git
-	@echo -e "$(GREEN)✓ Python environment ready$(NC)"
-
-$(VENV)/bin/activate:
-	@echo -e "$(BLUE)Creating Python virtual environment...$(NC)"
-	@PY_MINOR=$$($(PYTHON) -c "import sys; print(sys.version_info[1])"); \
-	if [ "$$PY_MINOR" -gt 13 ]; then \
-		echo -e "$(YELLOW)Warning: Python 3.$$PY_MINOR detected. ML packages may not be compatible.$(NC)"; \
-		echo -e "$(YELLOW)Recommended: Use Python 3.12 or 3.13 (brew install [email protected])$(NC)"; \
-	fi
-	$(PYTHON) -m venv $(VENV)
-
-setup-rust: ## Install Rust toolchain (if not present)
-	@command -v rustc >/dev/null 2>&1 || curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
-
-# =============================================================================
-# DEVELOPMENT
-# =============================================================================
-
-.PHONY: dev dev-backend dev-frontend dev-web kill-dev
-
-dev: ## Start backend + desktop app (parallel)
-	@echo -e "$(BLUE)Starting development servers...$(NC)"
-	@echo -e "$(YELLOW)Note: If Tauri fails, run 'make build-server' first or use separate terminals$(NC)"
-	@trap 'kill 0' EXIT; \
-		$(MAKE) dev-backend & \
-		sleep 2 && if [ "$$(uname)" = "Linux" ] && lspci 2>/dev/null | grep -qi nvidia; then \
-			WEBKIT_DISABLE_DMABUF_RENDERER=1 $(MAKE) dev-frontend; \
-		else \
-			$(MAKE) dev-frontend; \
-		fi & \
-		wait
-
-dev-backend: ## Start FastAPI backend server
-	@echo -e "$(BLUE)Starting backend server on http://localhost:17493$(NC)"
-	$(VENV_BIN)/uvicorn backend.main:app --reload --port 17493
-
-dev-frontend: ## Start Tauri desktop app
-	@echo -e "$(BLUE)Starting Tauri desktop app...$(NC)"
-	bun run dev
-
-dev-web: ## Start backend + web app (parallel)
-	@echo -e "$(BLUE)Starting web development servers...$(NC)"
-	@trap 'kill 0' EXIT; \
-		$(MAKE) dev-backend & \
-		sleep 2 && cd $(WEB_DIR) && bun run dev & \
-		wait
-
-kill-dev: ## Kill all development processes
-	@echo -e "$(YELLOW)Killing development processes...$(NC)"
-	-pkill -f "uvicorn main:app" 2>/dev/null || true
-	-pkill -f "vite" 2>/dev/null || true
-	@echo -e "$(GREEN)✓ Processes killed$(NC)"
-
-# =============================================================================
-# BUILD
-# =============================================================================
-
-.PHONY: build build-server build-tauri build-web
-
-build: build-server build-tauri ## Build everything (server binary + desktop app)
-	@echo -e "$(GREEN)✓ Build complete!$(NC)"
-
-build-server: ## Build Python server binary
-	@echo -e "$(BLUE)Building server binary...$(NC)"
-	PATH="$(VENV_BIN):$$PATH" ./scripts/build-server.sh
-
-build-tauri: ## Build Tauri desktop app
-	@echo -e "$(BLUE)Building Tauri desktop app...$(NC)"
-	cd $(TAURI_DIR) && bun run tauri build
-
-build-web: ## Build web app
-	@echo -e "$(BLUE)Building web app...$(NC)"
-	cd $(WEB_DIR) && bun run build
-	@echo -e "$(GREEN)✓ Web build output in $(WEB_DIR)/dist/$(NC)"
-
-# =============================================================================
-# DATABASE & API
-# =============================================================================
-
-.PHONY: db-init db-reset generate-api
-
-db-init: $(VENV)/bin/activate ## Initialize SQLite database
-	@echo -e "$(BLUE)Initializing database...$(NC)"
-	cd $(BACKEND_DIR) && $(PYTHON_VENV) -c "from database import init_db; init_db()"
-	@echo -e "$(GREEN)✓ Database created at $(BACKEND_DIR)/data/voicebox.db$(NC)"
-
-db-reset: ## Reset database (delete and reinitialize)
-	@echo -e "$(YELLOW)Resetting database...$(NC)"
-	rm -f $(BACKEND_DIR)/data/voicebox.db
-	$(MAKE) db-init
-
-generate-api: ## Generate TypeScript API client from OpenAPI schema
-	@echo -e "$(BLUE)Generating API client...$(NC)"
-	@echo -e "$(YELLOW)Note: Backend must be running (make dev-backend)$(NC)"
-	./scripts/generate-api.sh
-	@echo -e "$(GREEN)✓ API client generated in $(APP_DIR)/src/lib/api/$(NC)"
-
-# =============================================================================
-# CODE QUALITY
-# =============================================================================
-
-.PHONY: lint format typecheck check
-
-lint: ## Run linter (Biome)
-	@echo -e "$(BLUE)Linting...$(NC)"
-	bun run lint
-
-format: ## Format code (Biome)
-	@echo -e "$(BLUE)Formatting...$(NC)"
-	bun run format
-
-typecheck: ## Run TypeScript type checking
-	@echo -e "$(BLUE)Type checking...$(NC)"
-	bun run tsc --noEmit
-
-check: ## Run all checks (Biome lint + format + type check)
-	@echo -e "$(BLUE)Running all checks...$(NC)"
-	bun run check
-	@echo -e "$(GREEN)✓ All checks passed$(NC)"
-
-# =============================================================================
-# TESTING
-# =============================================================================
-
-.PHONY: test test-backend test-frontend
-
-test: test-backend test-frontend ## Run all tests
-	@echo -e "$(GREEN)✓ All tests passed$(NC)"
-
-test-backend: ## Run Python backend tests (requires pytest)
-	@echo -e "$(BLUE)Running backend tests...$(NC)"
-	@if [ -f "$(VENV_BIN)/pytest" ]; then \
-		cd $(BACKEND_DIR) && $(VENV_BIN)/pytest -v; \
-	else \
-		echo -e "$(YELLOW)pytest not installed. Run: $(PIP) install pytest$(NC)"; \
-		exit 1; \
-	fi
-
-test-frontend: ## Run frontend tests (requires test script in package.json)
-	@echo -e "$(BLUE)Running frontend tests...$(NC)"
-	@if bun run test --help >/dev/null 2>&1; then \
-		bun run test; \
-	else \
-		echo -e "$(YELLOW)No test script configured$(NC)"; \
-		exit 1; \
-	fi
-
-# =============================================================================
-# LOGS & DEBUGGING
-# =============================================================================
-
-.PHONY: logs docs
-
-logs: ## Tail backend logs
-	@echo -e "$(BLUE)Tailing logs (Ctrl+C to stop)...$(NC)"
-	tail -f $(BACKEND_DIR)/logs/*.log 2>/dev/null || echo "No log files found"
-
-docs: ## Open API documentation (backend must be running)
-	@echo -e "$(BLUE)Opening API docs...$(NC)"
-	open http://localhost:17493/docs 2>/dev/null || xdg-open http://localhost:17493/docs
-
-# =============================================================================
-# CLEAN
-# =============================================================================
-
-.PHONY: clean clean-python clean-build clean-all
-
-clean: ## Clean build artifacts
-	@echo -e "$(BLUE)Cleaning build artifacts...$(NC)"
-	rm -rf $(TAURI_DIR)/src-tauri/target/release
-	rm -rf $(WEB_DIR)/dist
-	rm -rf $(APP_DIR)/dist
-	@echo -e "$(GREEN)✓ Build artifacts cleaned$(NC)"
-
-clean-python: ## Clean Python cache and virtual environment
-	@echo -e "$(BLUE)Cleaning Python files...$(NC)"
-	rm -rf $(VENV)
-	find $(BACKEND_DIR) -type d -name "__pycache__" -exec rm -rf {} + 2>/dev/null || true
-	find $(BACKEND_DIR) -type f -name "*.pyc" -delete 2>/dev/null || true
-	@echo -e "$(GREEN)✓ Python environment cleaned$(NC)"
-
-clean-build: ## Clean Rust/Tauri build cache
-	@echo -e "$(BLUE)Cleaning Rust build cache...$(NC)"
-	cd $(TAURI_DIR)/src-tauri && cargo clean
-	@echo -e "$(GREEN)✓ Rust cache cleaned$(NC)"
-
-clean-all: clean clean-python clean-build ## Nuclear clean (everything)
-	@echo -e "$(BLUE)Cleaning node_modules...$(NC)"
-	rm -rf node_modules
-	rm -rf $(APP_DIR)/node_modules
-	rm -rf $(TAURI_DIR)/node_modules
-	rm -rf $(WEB_DIR)/node_modules
-	@echo -e "$(GREEN)✓ Full clean complete$(NC)"
@@ -31,7 +31,7 @@ Two-part fix:

 ## Testing
 To test this fix:
-1. Build Voicebox from source: `make build`
+1. Build Voicebox from source: `just build`
 2. Disconnect from internet
 3. Try generating speech
 4. Should work without network requests
@@ -40,13 +40,13 @@ To test this fix:

 ```bash
 # Install dependencies
-pip install -r requirements.txt
+just setup

 # Build the app
-make build
+just build

 # Or build just the server
-make build-server
+just build-server
 ```

 ## Notes
@@ -6,7 +6,7 @@

 <p align="center">
  <strong>The open-source voice synthesis studio.</strong><br/>
-  Clone voices. Generate speech. Build voice-powered apps.<br/>
+  Clone voices. Generate speech. Apply effects. Build voice-powered apps.<br/>
  All running locally on your machine.
 </p>

@@ -59,96 +59,147 @@

 ## What is Voicebox?

-Voicebox is a **local-first voice cloning studio** with DAW-like features for professional voice synthesis. Think of it as a **local, free and open-source alternative to ElevenLabs** — download models, clone voices, and generate speech entirely on your machine.
-
-Unlike cloud services that lock your voice data behind subscriptions, Voicebox gives you:
+Voicebox is a **local-first voice cloning studio** — a free and open-source alternative to ElevenLabs. Clone voices from a few seconds of audio, generate speech in 23 languages across 4 TTS engines, apply post-processing effects, and compose multi-voice projects with a timeline editor.

 - **Complete privacy** — models and voice data stay on your machine
- **Professional tools** — multi-track timeline editor, audio trimming, conversation mixing
- **Model flexibility** — currently powered by Qwen3-TTS, with support for XTTS, Bark, and other models coming soon
- **API-first** — use the desktop app or integrate voice synthesis into your own projects
+- **4 TTS engines** — Qwen3-TTS, LuxTTS, Chatterbox Multilingual, and Chatterbox Turbo
+- **23 languages** — from English to Arabic, Japanese, Hindi, Swahili, and more
+- **Post-processing effects** — pitch shift, reverb, delay, chorus, compression, and filters
+- **Expressive speech** — paralinguistic tags like `[laugh]`, `[sigh]`, `[gasp]` via Chatterbox Turbo
+- **Unlimited length** — auto-chunking with crossfade for scripts, articles, and chapters
+- **Stories editor** — multi-track timeline for conversations, podcasts, and narratives
+- **API-first** — REST API for integrating voice synthesis into your own projects
 - **Native performance** — built with Tauri (Rust), not Electron
- **Super fast on Mac** — MLX backend with native Metal acceleration for 4-5x faster inference on Apple Silicon
-
-Download a voice model, clone any voice from a few seconds of audio, and compose multi-voice projects with studio-grade editing tools. No Python install required, no cloud dependency, no limits.
+- **Runs everywhere** — macOS (MLX/Metal), Windows (CUDA), Linux, AMD ROCm, Intel Arc, Docker

 ---

 ## Download

-Voicebox is available now for macOS and Windows.
-
 | Platform | Download |
 |----------|----------|
-| macOS (Apple Silicon) | [Voicebox_aarch64.app.tar.gz](https://github.com/jamiepine/voicebox/releases/latest/download/Voicebox_aarch64.app.tar.gz) |
-| macOS (Intel) | [Voicebox_x64.app.tar.gz](https://github.com/jamiepine/voicebox/releases/latest/download/Voicebox_x64.app.tar.gz) |
-| Windows (MSI) | [Latest Windows MSI](https://github.com/jamiepine/voicebox/releases/latest) |
-| Windows (Setup) | [Latest Windows Setup](https://github.com/jamiepine/voicebox/releases/latest) |
+| macOS (Apple Silicon) | [Download DMG](https://voicebox.sh/download/mac-arm) |
+| macOS (Intel) | [Download DMG](https://voicebox.sh/download/mac-intel) |
+| Windows | [Download MSI](https://voicebox.sh/download/windows) |
+| Docker | `docker compose up` |

-> **Linux** — Pre-built binaries are not yet available. Linux users can compile from source, see [Development](#development) below.
+> **[View all binaries →](https://github.com/jamiepine/voicebox/releases/latest)**
+
+> **Linux** — Pre-built binaries are not yet available. See [voicebox.sh/linux-install](https://voicebox.sh/linux-install) for build-from-source instructions.

 ---

 ## Features

-### Voice Cloning with Qwen3-TTS
+### Multi-Engine Voice Cloning

-Powered by Alibaba's **Qwen3-TTS** — a breakthrough model that achieves near-perfect voice cloning from just a few seconds of audio.
+Four TTS engines with different strengths, switchable per-generation:

- **Instant cloning** — Upload a sample, get a voice profile
- **High fidelity** — Natural prosody, emotion, and cadence
- **Multi-language** — English, Chinese, and more coming
- **Lightning fast on Mac** — MLX backend leverages Apple Silicon's Neural Engine for super-fast generation
+| Engine | Languages | Strengths |
+|--------|-----------|-----------|
+| **Qwen3-TTS** (0.6B / 1.7B) | 10 | High-quality multilingual cloning, delivery instructions ("speak slowly", "whisper") |
+| **LuxTTS** | English | Lightweight (~1GB VRAM), 48kHz output, 150x realtime on CPU |
+| **Chatterbox Multilingual** | 23 | Broadest language coverage — Arabic, Danish, Finnish, Greek, Hebrew, Hindi, Malay, Norwegian, Polish, Swahili, Swedish, Turkish and more |
+| **Chatterbox Turbo** | English | Fast 350M model with paralinguistic emotion/sound tags |
+
+### Emotions & Paralinguistic Tags
+
+Type `/` in the text input to insert expressive tags that the model synthesizes inline with speech (Chatterbox Turbo):
+
+`[laugh]` `[chuckle]` `[gasp]` `[cough]` `[sigh]` `[groan]` `[sniff]` `[shush]` `[clear throat]`
+
+### Post-Processing Effects
+
+8 audio effects powered by Spotify's `pedalboard` library. Apply after generation, preview in real time, build reusable presets.
+
+| Effect | Description |
+|--------|-------------|
+| Pitch Shift | Up or down by up to 12 semitones |
+| Reverb | Configurable room size, damping, wet/dry mix |
+| Delay | Echo with adjustable time, feedback, and mix |
+| Chorus / Flanger | Modulated delay for metallic or lush textures |
+| Compressor | Dynamic range compression |
+| Gain | Volume adjustment (-40 to +40 dB) |
+| High-Pass Filter | Remove low frequencies |
+| Low-Pass Filter | Remove high frequencies |
+
+Ships with 4 built-in presets (Robotic, Radio, Echo Chamber, Deep Voice) and supports custom presets. Effects can be assigned per-profile as defaults.
+
+### Unlimited Generation Length
+
+Text is automatically split at sentence boundaries and each chunk is generated independently, then crossfaded together. Works with all engines.
+
+- Configurable auto-chunking limit (100–5,000 chars)
+- Crossfade slider (0–200ms) for smooth transitions
+- Max text length: 50,000 characters
+- Smart splitting respects abbreviations, CJK punctuation, and `[tags]`
+
+### Generation Versions
+
+Every generation supports multiple versions with provenance tracking:
+
+- **Original** — clean TTS output, always preserved
+- **Effects versions** — apply different effects chains from any source version
+- **Takes** — regenerate with a new seed for variation
+- **Source tracking** — each version records its lineage
+- **Favorites** — star generations for quick access
+
+### Async Generation Queue
+
+Generation is non-blocking. Submit and immediately start typing the next one.
+
+- Serial execution queue prevents GPU contention
+- Real-time SSE status streaming
+- Failed generations can be retried
+- Stale generations from crashes auto-recover on startup

 ### Voice Profile Management

- **Create profiles** from audio files or record directly in-app
- **Import/Export** profiles to share or back up
- **Multi-sample support** — combine multiple samples for higher quality cloning
- **Organize** with descriptions and language tags
-
-### Speech Generation
-
- **Text-to-speech** with any cloned voice
- **Batch generation** for long-form content
- **Smart caching** — regenerate instantly with voice prompt caching
+- Create profiles from audio files or record directly in-app
+- Import/export profiles to share or back up
+- Multi-sample support for higher quality cloning
+- Per-profile default effects chains
+- Organize with descriptions and language tags

 ### Stories Editor

-Create multi-voice narratives, podcasts, and conversations with a timeline-based editor.
+Multi-voice timeline editor for conversations, podcasts, and narratives.

- **Multi-track composition** — arrange multiple voice tracks in a single project
- **Inline audio editing** — trim and split clips directly in the timeline
- **Auto-playback** — preview stories with synchronized playhead
- **Voice mixing** — build conversations with multiple participants
+- Multi-track composition with drag-and-drop
+- Inline audio trimming and splitting
+- Auto-playback with synchronized playhead
+- Version pinning per track clip

 ### Recording & Transcription

- **In-app recording** with waveform visualization
- **System audio capture** — record desktop audio on macOS and Windows
- **Automatic transcription** powered by Whisper
- **Export recordings** in multiple formats
+- In-app recording with waveform visualization
+- System audio capture (macOS and Windows)
+- Automatic transcription powered by Whisper (including Whisper Turbo)
+- Export recordings in multiple formats

-### Generation History
+### Model Management

- **Full history** of all generated audio
- **Search & filter** by voice, text, or date
- **Re-generate** any past generation with one click
+- Per-model unload to free GPU memory without deleting downloads
+- Custom models directory via `VOICEBOX_MODELS_DIR`
+- Model folder migration with progress tracking
+- Download cancel/clear UI

-### Flexible Deployment
+### GPU Support

- **Local mode** — Everything runs on your machine
- **Remote mode** — Connect to a GPU server on your network
- **One-click server** — Turn any machine into a Voicebox server
+| Platform | Backend | Notes |
+|----------|---------|-------|
+| macOS (Apple Silicon) | MLX (Metal) | 4-5x faster via Neural Engine |
+| Windows / Linux (NVIDIA) | PyTorch (CUDA) | Auto-downloads CUDA binary from within the app |
+| Linux (AMD) | PyTorch (ROCm) | Auto-configures HSA_OVERRIDE_GFX_VERSION |
+| Windows (any GPU) | DirectML | Universal Windows GPU support |
+| Intel Arc | IPEX/XPU | Intel discrete GPU acceleration |
+| Any | CPU | Works everywhere, just slower |

 ---

 ## API

-Voicebox exposes a full REST API, so you can integrate voice synthesis into your own apps.
-
-For the current local app and development workflow, the backend is typically available at `http://localhost:17493`.
-If you launch the backend manually with a different host or port, use that address instead.
+Voicebox exposes a full REST API for integrating voice synthesis into your own apps.

 ```bash
 # Generate speech
@@ -165,15 +216,9 @@ curl -X POST http://localhost:17493/profiles \
  -d '{"name": "My Voice", "language": "en"}'
 ```

-**Use cases:**
+**Use cases:** game dialogue, podcast production, accessibility tools, voice assistants, content automation.

- Game dialogue systems
- Podcast/video production pipelines
- Accessibility tools
- Voice assistants
- Content creation automation
-
-Full API documentation is available at `http://localhost:17493/docs` in the default local workflow, or at `/docs` on whatever server address you configured.
+Full API documentation available at `http://localhost:17493/docs`.

 ---

@@ -185,42 +230,24 @@ Full API documentation is available at `http://localhost:17493/docs` in the defa
 | Frontend | React, TypeScript, Tailwind CSS |
 | State | Zustand, React Query |
 | Backend | FastAPI (Python) |
-| Voice Model | Qwen3-TTS (PyTorch or MLX) |
-| Transcription | Whisper (PyTorch or MLX) |
-| Inference Engine | MLX (Apple Silicon) / PyTorch (Windows/Linux/Intel) |
+| TTS Engines | Qwen3-TTS, LuxTTS, Chatterbox, Chatterbox Turbo |
+| Effects | Pedalboard (Spotify) |
+| Transcription | Whisper / Whisper Turbo (PyTorch or MLX) |
+| Inference | MLX (Apple Silicon) / PyTorch (CUDA/ROCm/XPU/CPU) |
 | Database | SQLite |
 | Audio | WaveSurfer.js, librosa |

-**Why this stack?**
-
- **Tauri over Electron** — 10x smaller bundle, native performance, lower memory
- **FastAPI** — Async Python with automatic OpenAPI schema generation
- **Type-safe end-to-end** — Generated TypeScript client from OpenAPI spec
-
 ---

 ## Roadmap

-Voicebox is the beginning of something bigger. Here's what's coming:
-
-### Coming Soon
-
 | Feature | Description |
 |---------|-------------|
-| **Real-time Synthesis** | Stream audio as it generates, word by word |
-| **Conversation Mode** | Multi-speaker dialogues with automatic turn-taking |
-| **Voice Effects** | Pitch shift, reverb, M3GAN-style effects |
-| **Timeline Editor** | Audio studio with word-level precision editing |
+| **Real-time Streaming** | Stream audio as it generates, word by word |
+| **Voice Design** | Create new voices from text descriptions |
 | **More Models** | XTTS, Bark, and other open-source voice models |
-
-### Future Vision
-
- **Voice Design** — Create new voices from text descriptions
- **Project System** — Save and load complex multi-voice sessions
- **Plugin Architecture** — Extend with custom models and effects
- **Mobile Companion** — Control Voicebox from your phone
-
-Voicebox aims to be the **one-stop shop for everything voice** — cloning, synthesis, editing, effects, and beyond.
+| **Plugin Architecture** | Extend with custom models and effects |
+| **Mobile Companion** | Control Voicebox from your phone |

 ---

@@ -242,14 +269,6 @@ Install [just](https://github.com/casey/just): `brew install just` or `cargo ins

 **Prerequisites:** [Bun](https://bun.sh), [Rust](https://rustup.rs), [Python 3.11+](https://python.org), [Tauri Prerequisites](https://v2.tauri.app/start/prerequisites/), and [Xcode](https://developer.apple.com/xcode/) on macOS.

-### Platform Notes
-
-| Platform | GPU Backend | Notes |
-|----------|-------------|-------|
-| macOS (Apple Silicon) | MLX (Metal) | 4-5x faster inference via Neural Engine |
-| Windows (NVIDIA) | PyTorch (CUDA) | `just setup` auto-installs CUDA PyTorch |
-| Windows/Linux (no NVIDIA) | PyTorch (CPU) | Works but slower |
-
 ### Building Locally

 ```bash
@@ -257,8 +276,6 @@ just build          # Build CPU server binary + Tauri app
 just build-local    # (Windows) Build CPU + CUDA server binaries + Tauri app
 ```

-`just build-local` produces a production-ready installer with the CUDA binary pre-placed for GPU switching.
-
 ### Project Structure

 ```
@@ -1,6 +1,6 @@
 {
  "name": "@voicebox/app",
-  "version": "0.2.1",
+  "version": "0.2.3",
  "private": true,
  "type": "module",
  "scripts": {
@@ -0,0 +1,103 @@
+import type { UseFormReturn } from 'react-hook-form';
+import { FormControl } from '@/components/ui/form';
+import {
+  Select,
+  SelectContent,
+  SelectItem,
+  SelectTrigger,
+  SelectValue,
+} from '@/components/ui/select';
+import { getLanguageOptionsForEngine } from '@/lib/constants/languages';
+import type { GenerationFormValues } from '@/lib/hooks/useGenerationForm';
+
+/**
+ * Engine/model options and their display metadata.
+ * Adding a new engine means adding one entry here.
+ */
+const ENGINE_OPTIONS = [
+  { value: 'qwen:1.7B', label: 'Qwen3-TTS 1.7B' },
+  { value: 'qwen:0.6B', label: 'Qwen3-TTS 0.6B' },
+  { value: 'luxtts', label: 'LuxTTS' },
+  { value: 'chatterbox', label: 'Chatterbox' },
+  { value: 'chatterbox_turbo', label: 'Chatterbox Turbo' },
+] as const;
+
+const ENGINE_DESCRIPTIONS: Record<string, string> = {
+  qwen: 'Multi-language, two sizes',
+  luxtts: 'Fast, English-focused',
+  chatterbox: '23 languages, incl. Hebrew',
+  chatterbox_turbo: 'English, [laugh] [cough] tags',
+};
+
+/** Engines that only support English and should force language to 'en' on select. */
+const ENGLISH_ONLY_ENGINES = new Set(['luxtts', 'chatterbox_turbo']);
+
+function getSelectValue(engine: string, modelSize?: string): string {
+  if (engine === 'qwen') return `qwen:${modelSize || '1.7B'}`;
+  return engine;
+}
+
+function handleEngineChange(form: UseFormReturn<GenerationFormValues>, value: string) {
+  if (value.startsWith('qwen:')) {
+    const [, modelSize] = value.split(':');
+    form.setValue('engine', 'qwen');
+    form.setValue('modelSize', modelSize as '1.7B' | '0.6B');
+    // Validate language is supported by Qwen
+    const currentLang = form.getValues('language');
+    const available = getLanguageOptionsForEngine('qwen');
+    if (!available.some((l) => l.value === currentLang)) {
+      form.setValue('language', available[0]?.value ?? 'en');
+    }
+  } else {
+    form.setValue('engine', value as GenerationFormValues['engine']);
+    form.setValue('modelSize', undefined as unknown as '1.7B' | '0.6B');
+    if (ENGLISH_ONLY_ENGINES.has(value)) {
+      form.setValue('language', 'en');
+    } else {
+      // If current language isn't supported by the new engine, reset to first available
+      const currentLang = form.getValues('language');
+      const available = getLanguageOptionsForEngine(value);
+      if (!available.some((l) => l.value === currentLang)) {
+        form.setValue('language', available[0]?.value ?? 'en');
+      }
+    }
+  }
+}
+
+interface EngineModelSelectorProps {
+  form: UseFormReturn<GenerationFormValues>;
+  compact?: boolean;
+}
+
+export function EngineModelSelector({ form, compact }: EngineModelSelectorProps) {
+  const engine = form.watch('engine') || 'qwen';
+  const modelSize = form.watch('modelSize');
+  const selectValue = getSelectValue(engine, modelSize);
+
+  const itemClass = compact ? 'text-xs text-muted-foreground' : undefined;
+  const triggerClass = compact
+    ? 'h-8 text-xs bg-card border-border rounded-full hover:bg-background/50 transition-all'
+    : undefined;
+
+  return (
+    <Select value={selectValue} onValueChange={(v) => handleEngineChange(form, v)}>
+      <FormControl>
+        <SelectTrigger className={triggerClass}>
+          <SelectValue />
+        </SelectTrigger>
+      </FormControl>
+      <SelectContent>
+        {ENGINE_OPTIONS.map((opt) => (
+          <SelectItem key={opt.value} value={opt.value} className={itemClass}>
+            {opt.label}
+          </SelectItem>
+        ))}
+      </SelectContent>
+    </Select>
+  );
+}
+
+/** Returns a human-readable description for the currently selected engine. */
+export function getEngineDescription(engine: string): string {
+  return ENGINE_DESCRIPTIONS[engine] ?? '';
+}
@@ -1,8 +1,8 @@
+import { useQuery } from '@tanstack/react-query';
 import { useMatchRoute } from '@tanstack/react-router';
 import { AnimatePresence, motion } from 'framer-motion';
-import { Loader2, SlidersHorizontal, Sparkles } from 'lucide-react';
+import { Loader2, Sparkles } from 'lucide-react';
 import { useEffect, useRef, useState } from 'react';
-import { EffectsChainEditor } from '@/components/Effects/EffectsChainEditor';
 import { Button } from '@/components/ui/button';
 import { Form, FormControl, FormField, FormItem, FormMessage } from '@/components/ui/form';
 import {
@@ -13,7 +13,7 @@ import {
  SelectValue,
 } from '@/components/ui/select';
 import { Textarea } from '@/components/ui/textarea';
-import type { EffectConfig } from '@/lib/api/types';
+import { apiClient } from '@/lib/api/client';
 import { getLanguageOptionsForEngine, type LanguageCode } from '@/lib/constants/languages';
 import { useGenerationForm } from '@/lib/hooks/useGenerationForm';
 import { useProfile, useProfiles } from '@/lib/hooks/useProfiles';
@@ -22,6 +22,7 @@ import { cn } from '@/lib/utils/cn';
 import { useGenerationStore } from '@/stores/generationStore';
 import { useStoryStore } from '@/stores/storyStore';
 import { useUIStore } from '@/stores/uiStore';
+import { EngineModelSelector } from './EngineModelSelector';
 import { ParalinguisticInput } from './ParalinguisticInput';

 interface FloatingGenerateBoxProps {
@@ -38,8 +39,7 @@ export function FloatingGenerateBox({
  const { data: selectedProfile } = useProfile(selectedProfileId || '');
  const { data: profiles } = useProfiles();
  const [isExpanded, setIsExpanded] = useState(false);
-  const [isInstructMode, setIsInstructMode] = useState(false);
-  const [effectsChain, setEffectsChain] = useState<EffectConfig[]>([]);
+  const [selectedPresetId, setSelectedPresetId] = useState<string | null>(null);
  const containerRef = useRef<HTMLDivElement>(null);
  const textareaRef = useRef<HTMLTextAreaElement | null>(null);
  const matchRoute = useMatchRoute();
@@ -49,18 +49,28 @@ export function FloatingGenerateBox({
  const { data: currentStory } = useStory(selectedStoryId);
  const addPendingStoryAdd = useGenerationStore((s) => s.addPendingStoryAdd);

+  // Fetch effect presets for the dropdown
+  const { data: effectPresets } = useQuery({
+    queryKey: ['effectPresets'],
+    queryFn: () => apiClient.listEffectPresets(),
+  });
+
  // Calculate if track editor is visible (on stories route with items)
  const hasTrackEditor = isStoriesRoute && currentStory && currentStory.items.length > 0;

  const { form, handleSubmit, isPending } = useGenerationForm({
    onSuccess: async (generationId) => {
      setIsExpanded(false);
-      // Defer the story add until TTS completes — useGenerationProgress handles it
+      // Defer the story add until TTS completes -- useGenerationProgress handles it
      if (isStoriesRoute && selectedStoryId && generationId) {
        addPendingStoryAdd(generationId, selectedStoryId);
      }
    },
-    getEffectsChain: () => (effectsChain.length > 0 ? effectsChain : undefined),
+    getEffectsChain: () => {
+      if (!selectedPresetId || !effectPresets) return undefined;
+      const preset = effectPresets.find((p) => p.id === selectedPresetId);
+      return preset?.effects_chain;
+    },
  });

  // Click away handler to collapse the box
@@ -188,111 +198,57 @@ export function FloatingGenerateBox({
        <Form {...form}>
          <form onSubmit={form.handleSubmit(onSubmit)}>
            <div className="flex gap-2">
-              <motion.div
-                className={cn('flex-1', isExpanded && 'mr-12')}
-                transition={{ duration: 0.3, ease: 'easeOut' }}
-              >
-                {/* Text field - hidden when in instruct mode */}
-                <div style={{ display: isInstructMode ? 'none' : 'block' }}>
-                  <FormField
-                    control={form.control}
-                    name="text"
-                    render={({ field }) => (
-                      <FormItem>
-                        <FormControl>
-                          <motion.div
-                            animate={{
-                              height: isExpanded ? 'auto' : '32px',
-                            }}
-                            transition={{ duration: 0.15, ease: 'easeOut' }}
-                            style={{ overflow: 'hidden' }}
-                          >
-                            {form.watch('engine') === 'chatterbox_turbo' ? (
-                              <ParalinguisticInput
-                                value={field.value}
-                                onChange={field.onChange}
-                                placeholder={
-                                  isStoriesRoute && currentStory
-                                    ? `Generate speech for "${currentStory.name}"... (type / for effects)`
-                                    : selectedProfile
-                                      ? `Type / for effects like [laugh], [sigh]...`
-                                      : 'Select a voice profile above...'
-                                }
-                                className="px-3 py-2 resize-none bg-transparent border-none focus-visible:ring-0 focus-visible:ring-offset-0 focus:outline-none focus:ring-0 outline-none ring-0 rounded-2xl text-sm w-full"
-                                style={{
-                                  minHeight: isExpanded ? '100px' : '32px',
-                                  maxHeight: '300px',
-                                  overflowY: 'auto',
-                                }}
-                                disabled={!selectedProfileId}
-                                onClick={() => setIsExpanded(true)}
-                                onFocus={() => setIsExpanded(true)}
-                              />
-                            ) : (
-                              <Textarea
-                                {...field}
-                                ref={(node: HTMLTextAreaElement | null) => {
-                                  // Store ref for auto-resize (only for active field)
-                                  if (!isInstructMode) {
-                                    textareaRef.current = node;
-                                  }
-                                  // Forward ref to react-hook-form
-                                  if (typeof field.ref === 'function') {
-                                    field.ref(node);
-                                  }
-                                }}
-                                placeholder={
-                                  isStoriesRoute && currentStory
-                                    ? `Generate speech for "${currentStory.name}"...`
-                                    : selectedProfile
-                                      ? `Generate speech using ${selectedProfile.name}...`
-                                      : 'Select a voice profile above...'
-                                }
-                                className="resize-none bg-transparent border-none focus-visible:ring-0 focus-visible:ring-offset-0 focus:outline-none focus:ring-0 outline-none ring-0 rounded-2xl text-sm placeholder:text-muted-foreground/60 w-full"
-                                style={{
-                                  minHeight: isExpanded ? '100px' : '32px',
-                                  maxHeight: '300px',
-                                }}
-                                disabled={!selectedProfileId}
-                                onClick={() => setIsExpanded(true)}
-                                onFocus={() => setIsExpanded(true)}
-                              />
-                            )}
-                          </motion.div>
-                        </FormControl>
-                        <FormMessage className="text-xs" />
-                      </FormItem>
-                    )}
-                  />
-                </div>
-                {/* Instruct field - hidden when in text mode */}
-                <div style={{ display: isInstructMode ? 'block' : 'none' }}>
-                  <FormField
-                    control={form.control}
-                    name="instruct"
-                    render={({ field }) => (
-                      <FormItem>
-                        <FormControl>
-                          <motion.div
-                            animate={{
-                              height: isExpanded ? 'auto' : '32px',
-                            }}
-                            transition={{ duration: 0.15, ease: 'easeOut' }}
-                            style={{ overflow: 'hidden' }}
-                          >
+              <motion.div className="flex-1" transition={{ duration: 0.3, ease: 'easeOut' }}>
+                <FormField
+                  control={form.control}
+                  name="text"
+                  render={({ field }) => (
+                    <FormItem>
+                      <FormControl>
+                        <motion.div
+                          animate={{
+                            height: isExpanded ? 'auto' : '32px',
+                          }}
+                          transition={{ duration: 0.15, ease: 'easeOut' }}
+                          style={{ overflow: 'hidden' }}
+                        >
+                          {form.watch('engine') === 'chatterbox_turbo' ? (
+                            <ParalinguisticInput
+                              value={field.value}
+                              onChange={field.onChange}
+                              placeholder={
+                                isStoriesRoute && currentStory
+                                  ? `Generate speech for "${currentStory.name}"... (type / for effects)`
+                                  : selectedProfile
+                                    ? `Type / for effects like [laugh], [sigh]...`
+                                    : 'Select a voice profile above...'
+                              }
+                              className="px-3 py-2 resize-none bg-transparent border-none focus-visible:ring-0 focus-visible:ring-offset-0 focus:outline-none focus:ring-0 outline-none ring-0 rounded-2xl text-sm w-full"
+                              style={{
+                                minHeight: isExpanded ? '100px' : '32px',
+                                maxHeight: '300px',
+                                overflowY: 'auto',
+                              }}
+                              disabled={!selectedProfileId}
+                              onClick={() => setIsExpanded(true)}
+                              onFocus={() => setIsExpanded(true)}
+                            />
+                          ) : (
                            <Textarea
                              {...field}
                              ref={(node: HTMLTextAreaElement | null) => {
-                                // Store ref for auto-resize (only for active field)
-                                if (isInstructMode) {
-                                  textareaRef.current = node;
-                                }
-                                // Forward ref to react-hook-form
+                                textareaRef.current = node;
                                if (typeof field.ref === 'function') {
                                  field.ref(node);
                                }
                              }}
-                              placeholder="e.g. very happy and excited"
+                              placeholder={
+                                isStoriesRoute && currentStory
+                                  ? `Generate speech for "${currentStory.name}"...`
+                                  : selectedProfile
+                                    ? `Generate speech using ${selectedProfile.name}...`
+                                    : 'Select a voice profile above...'
+                              }
                              className="resize-none bg-transparent border-none focus-visible:ring-0 focus-visible:ring-offset-0 focus:outline-none focus:ring-0 outline-none ring-0 rounded-2xl text-sm placeholder:text-muted-foreground/60 w-full"
                              style={{
                                minHeight: isExpanded ? '100px' : '32px',
@@ -302,13 +258,13 @@ export function FloatingGenerateBox({
                              onClick={() => setIsExpanded(true)}
                              onFocus={() => setIsExpanded(true)}
                            />
-                          </motion.div>
-                        </FormControl>
-                        <FormMessage className="text-xs" />
-                      </FormItem>
-                    )}
-                  />
-                </div>
+                          )}
+                        </motion.div>
+                      </FormControl>
+                      <FormMessage className="text-xs" />
+                    </FormItem>
+                  )}
+                />
              </motion.div>

              <div className="relative shrink-0">
@@ -340,62 +296,9 @@ export function FloatingGenerateBox({
                        : 'Generate speech'}
                  </span>
                </div>
-                <AnimatePresence>
-                  {isExpanded && form.watch('engine') === 'qwen' && (
-                    <motion.div
-                      initial={{ opacity: 0, scale: 0.8 }}
-                      animate={{ opacity: 1, scale: 1 }}
-                      exit={{ opacity: 0, scale: 0.8 }}
-                      transition={{ duration: 0.2 }}
-                      className="absolute top-0 right-[calc(100%+0.5rem)]"
-                    >
-                      <div className="group relative">
-                        <Button
-                          type="button"
-                          variant="ghost"
-                          size="icon"
-                          onClick={() => setIsInstructMode(!isInstructMode)}
-                          className={cn(
-                            'h-10 w-10 rounded-full transition-all duration-200',
-                            isInstructMode
-                              ? 'bg-accent text-accent-foreground border border-accent hover:bg-accent/90'
-                              : effectsChain.length > 0
-                                ? 'bg-accent/50 text-accent-foreground border border-accent/50 hover:bg-accent/70'
-                                : 'bg-card border border-border hover:bg-background/50',
-                          )}
-                          aria-label={
-                            isInstructMode ? 'Fine tune instructions, on' : 'Fine tune instructions'
-                          }
-                        >
-                          <SlidersHorizontal className="h-4 w-4" />
-                        </Button>
-                        <span className="pointer-events-none absolute bottom-full left-1/2 -translate-x-1/2 mb-2 whitespace-nowrap rounded-md bg-popover px-3 py-1.5 text-xs text-popover-foreground border border-border opacity-0 transition-opacity group-hover:opacity-100 z-[9999]">
-                          Fine tune instructions & effects
-                        </span>
-                      </div>
-                    </motion.div>
-                  )}
-                </AnimatePresence>
              </div>
            </div>

-            {/* Effects chain editor panel - shown alongside instruct */}
-            <AnimatePresence>
-              {isExpanded && isInstructMode && (
-                <motion.div
-                  initial={{ height: 0, opacity: 0 }}
-                  animate={{ height: 'auto', opacity: 1 }}
-                  exit={{ height: 0, opacity: 0 }}
-                  transition={{ duration: 0.2 }}
-                  className="overflow-hidden mt-2"
-                >
-                  <div className="border-t border-border/50 pt-2 pb-1">
-                    <EffectsChainEditor value={effectsChain} onChange={setEffectsChain} compact />
-                  </div>
-                </motion.div>
-              )}
-            </AnimatePresence>
-
            <AnimatePresence>
              <motion.div
                initial={{ height: 0, opacity: 0 }}
@@ -454,57 +357,29 @@ export function FloatingGenerateBox({
                    }}
                  />

+                  <FormItem className="flex-1 space-y-0">
+                    <EngineModelSelector form={form} compact />
+                  </FormItem>
+
                  <FormItem className="flex-1 space-y-0">
                    <Select
-                      value={
-                        form.watch('engine') === 'luxtts'
-                          ? 'luxtts'
-                          : form.watch('engine') === 'chatterbox'
-                            ? 'chatterbox'
-                            : form.watch('engine') === 'chatterbox_turbo'
-                              ? 'chatterbox_turbo'
-                              : `qwen:${form.watch('modelSize') || '1.7B'}`
+                      value={selectedPresetId || 'none'}
+                      onValueChange={(value) =>
+                        setSelectedPresetId(value === 'none' ? null : value)
                      }
-                      onValueChange={(value) => {
-                        if (value === 'luxtts') {
-                          form.setValue('engine', 'luxtts');
-                          form.setValue('language', 'en');
-                        } else if (value === 'chatterbox') {
-                          form.setValue('engine', 'chatterbox');
-                        } else if (value === 'chatterbox_turbo') {
-                          form.setValue('engine', 'chatterbox_turbo');
-                          form.setValue('language', 'en');
-                        } else {
-                          const [, modelSize] = value.split(':');
-                          form.setValue('engine', 'qwen');
-                          form.setValue('modelSize', modelSize as '1.7B' | '0.6B');
-                        }
-                      }}
                    >
-                      <FormControl>
-                        <SelectTrigger className="h-8 text-xs bg-card border-border rounded-full hover:bg-background/50 transition-all">
-                          <SelectValue />
-                        </SelectTrigger>
-                      </FormControl>
+                      <SelectTrigger className="h-8 text-xs bg-card border-border rounded-full hover:bg-background/50 transition-all">
+                        <SelectValue placeholder="No effects" />
+                      </SelectTrigger>
                      <SelectContent>
-                        <SelectItem value="qwen:1.7B" className="text-xs text-muted-foreground">
-                          Qwen3-TTS 1.7B
-                        </SelectItem>
-                        <SelectItem value="qwen:0.6B" className="text-xs text-muted-foreground">
-                          Qwen3-TTS 0.6B
-                        </SelectItem>
-                        <SelectItem value="luxtts" className="text-xs text-muted-foreground">
-                          LuxTTS
-                        </SelectItem>
-                        <SelectItem value="chatterbox" className="text-xs text-muted-foreground">
-                          Chatterbox
-                        </SelectItem>
-                        <SelectItem
-                          value="chatterbox_turbo"
-                          className="text-xs text-muted-foreground"
-                        >
-                          Chatterbox Turbo
+                        <SelectItem value="none" className="text-xs">
+                          No effects
                        </SelectItem>
+                        {effectPresets?.map((preset) => (
+                          <SelectItem key={preset.id} value={preset.id} className="text-xs">
+                            {preset.name}
+                          </SelectItem>
+                        ))}
                      </SelectContent>
                    </Select>
                  </FormItem>
@@ -23,6 +23,7 @@ import { getLanguageOptionsForEngine } from '@/lib/constants/languages';
 import { useGenerationForm } from '@/lib/hooks/useGenerationForm';
 import { useProfile } from '@/lib/hooks/useProfiles';
 import { useUIStore } from '@/stores/uiStore';
+import { EngineModelSelector, getEngineDescription } from './EngineModelSelector';
 import { ParalinguisticInput } from './ParalinguisticInput';

 export function GenerationForm() {
@@ -117,53 +118,9 @@ export function GenerationForm() {
            <div className="grid gap-4 md:grid-cols-3">
              <FormItem>
                <FormLabel>Model</FormLabel>
-                <Select
-                  value={
-                    form.watch('engine') === 'luxtts'
-                      ? 'luxtts'
-                      : form.watch('engine') === 'chatterbox'
-                        ? 'chatterbox'
-                        : form.watch('engine') === 'chatterbox_turbo'
-                          ? 'chatterbox_turbo'
-                          : `qwen:${form.watch('modelSize') || '1.7B'}`
-                  }
-                  onValueChange={(value) => {
-                    if (value === 'luxtts') {
-                      form.setValue('engine', 'luxtts');
-                      form.setValue('language', 'en');
-                    } else if (value === 'chatterbox') {
-                      form.setValue('engine', 'chatterbox');
-                    } else if (value === 'chatterbox_turbo') {
-                      form.setValue('engine', 'chatterbox_turbo');
-                      form.setValue('language', 'en');
-                    } else {
-                      const [, modelSize] = value.split(':');
-                      form.setValue('engine', 'qwen');
-                      form.setValue('modelSize', modelSize as '1.7B' | '0.6B');
-                    }
-                  }}
-                >
-                  <FormControl>
-                    <SelectTrigger>
-                      <SelectValue />
-                    </SelectTrigger>
-                  </FormControl>
-                  <SelectContent>
-                    <SelectItem value="qwen:1.7B">Qwen3-TTS 1.7B</SelectItem>
-                    <SelectItem value="qwen:0.6B">Qwen3-TTS 0.6B</SelectItem>
-                    <SelectItem value="luxtts">LuxTTS</SelectItem>
-                    <SelectItem value="chatterbox">Chatterbox</SelectItem>
-                    <SelectItem value="chatterbox_turbo">Chatterbox Turbo</SelectItem>
-                  </SelectContent>
-                </Select>
+                <EngineModelSelector form={form} />
                <FormDescription>
-                  {form.watch('engine') === 'luxtts'
-                    ? 'Fast, English-focused'
-                    : form.watch('engine') === 'chatterbox'
-                      ? '23 languages, incl. Hebrew'
-                      : form.watch('engine') === 'chatterbox_turbo'
-                        ? 'English, [laugh] [cough] tags'
-                        : 'Multi-language, two sizes'}
+                  {getEngineDescription(form.watch('engine') || 'qwen')}
                </FormDescription>
              </FormItem>

@@ -394,7 +394,8 @@ export function HistoryTable() {
          >
            {history.map((gen) => {
              const isCurrentlyPlaying = currentAudioId === gen.id && isPlaying;
-              const isGenerating = gen.status === 'generating';
+              const isInProgress = gen.status === 'loading_model' || gen.status === 'generating';
+              const isGenerating = isInProgress;
              const isFailed = gen.status === 'failed';
              const isPlayable = !isGenerating && !isFailed;
              const hasVersions = gen.versions && gen.versions.length > 1;
@@ -472,8 +473,10 @@ export function HistoryTable() {
                        ) : null}
                      </div>
                      <div className="text-xs text-muted-foreground">
-                        {isGenerating ? (
-                          <span className="text-accent">Generating...</span>
+                        {isInProgress ? (
+                          <span className="text-accent">
+                            {gen.status === 'loading_model' ? 'Loading model...' : 'Generating...'}
+                          </span>
                        ) : (
                          formatDate(gen.created_at)
                        )}
@@ -246,13 +246,18 @@ export function GpuAcceleration() {
        {/* CUDA download section - only show when no GPU is active (native or CUDA) */}
        {!hasNativeGpu && !isCurrentlyCuda && (
          <>
-            {/* Download progress */}
+            {/* Download progress (manual download or auto-update) */}
            {cudaDownloading && downloadProgress && (
              <div className="space-y-2">
                <div className="flex items-center justify-between text-sm">
                  <div className="flex items-center gap-2">
                    <Loader2 className="h-4 w-4 animate-spin" />
-                    <span>{downloadProgress.filename || 'Downloading CUDA backend...'}</span>
+                    <span>
+                      {downloadProgress.filename ||
+                        (cudaAvailable
+                          ? 'Updating CUDA backend...'
+                          : 'Downloading CUDA backend...')}
+                    </span>
                  </div>
                  {downloadProgress.total > 0 && (
                    <span className="text-muted-foreground">
@@ -1,7 +1,10 @@
 import { Link, useMatchRoute } from '@tanstack/react-router';
 import { AudioLines, Box, Mic, Server, Speaker, Volume2, Wand2 } from 'lucide-react';
+import { useEffect, useState } from 'react';
 import voiceboxLogo from '@/assets/voicebox-logo.png';
 import { cn } from '@/lib/utils/cn';
+import { usePlatform } from '@/platform/PlatformContext';
+import type { UpdateStatus } from '@/platform/types';
 import { usePlayerStore } from '@/stores/playerStore';
 import { version } from '../../package.json';

@@ -22,6 +25,10 @@ const tabs = [
 export function Sidebar({ isMacOS }: SidebarProps) {
  const matchRoute = useMatchRoute();
  const isPlayerOpen = !!usePlayerStore((s) => s.audioUrl);
+  const platform = usePlatform();
+
+  const [updateStatus, setUpdateStatus] = useState<UpdateStatus>(platform.updater.getStatus());
+  useEffect(() => platform.updater.subscribe(setUpdateStatus), [platform.updater]);

  return (
    <div
@@ -85,10 +92,18 @@ export function Sidebar({ isMacOS }: SidebarProps) {

      {/* Version */}
      <div
-        className="mt-auto text-[10px] text-muted-foreground/50 transition-all duration-300"
+        className="mt-auto flex flex-col items-center gap-1.5 transition-all duration-300"
        style={{ paddingBottom: isPlayerOpen ? '7rem' : undefined }}
      >
-        v{version}
+        <span className="text-[10px] text-muted-foreground/50">v{version}</span>
+        {updateStatus.available && (
+          <Link
+            to="/server"
+            className="text-[9px] font-semibold tracking-wide uppercase px-2 py-0.5 rounded-full bg-accent/15 text-accent hover:bg-accent/25 transition-colors"
+          >
+            Update
+          </Link>
+        )}
      </div>
    </div>
  );
@@ -16,7 +16,7 @@ const SelectTrigger = React.forwardRef<
  <SelectPrimitive.Trigger
    ref={ref}
    className={cn(
-      'flex h-10 w-full items-center justify-between rounded-md border border-input bg-background px-3 py-2 text-sm ring-offset-background placeholder:text-muted-foreground focus:outline-none focus:ring-2 focus:ring-ring focus:ring-offset-2 disabled:cursor-not-allowed disabled:opacity-50 [&>span]:line-clamp-1',
+      'flex h-10 w-full items-center justify-between rounded-md border border-input bg-background px-3 py-2 text-sm ring-offset-background placeholder:text-muted-foreground focus:outline-none focus:bg-muted/50 disabled:cursor-not-allowed disabled:opacity-50 [&>span]:line-clamp-1',
      className,
    )}
    {...props}
@@ -73,7 +73,7 @@ export interface GenerationResponse {
  instruct?: string;
  engine?: string;
  model_size?: string;
-  status: 'generating' | 'completed' | 'failed';
+  status: 'loading_model' | 'generating' | 'completed' | 'failed';
  error?: string;
  is_favorited?: boolean;
  created_at: string;
@@ -8,7 +8,7 @@ import { useServerStore } from '@/stores/serverStore';

 interface GenerationStatusEvent {
  id: string;
-  status: 'generating' | 'completed' | 'failed' | 'not_found';
+  status: 'loading_model' | 'generating' | 'completed' | 'failed' | 'not_found';
  duration?: number;
  error?: string;
 }
@@ -1,462 +1,135 @@
-# voicebox Backend
+# Voicebox Backend

-Production-quality FastAPI backend for Qwen3-TTS voice cloning.
+FastAPI server powering voice cloning, speech generation, and audio processing. Runs locally as a Tauri sidecar or standalone via `python -m backend.main`.

-## Features
+## Running

- ✅ **Voice Profile Management** - Create, update, delete voice profiles with multi-sample support
- ✅ **Voice Cloning** - Generate speech using voice profiles with caching
- ✅ **Generation History** - Full history tracking with search and filtering
- ✅ **Transcription** - Whisper-based audio transcription
- ✅ **Multi-Sample Profiles** - Combine multiple reference samples for better quality
- ✅ **Voice Prompt Caching** - Dual memory + disk caching for fast generation
- ✅ **Audio Validation** - Automatic validation of reference audio quality
- ✅ **Model Management** - Lazy loading and VRAM management
+```bash
+# Via justfile (recommended)
+just dev:server
+
+# Standalone
+python -m backend.main --host 127.0.0.1 --port 17493
+
+# With custom data directory
+python -m backend.main --data-dir /path/to/data
+```
+
+The server auto-initializes the SQLite database on first startup. Models are downloaded from HuggingFace on first use.

 ## Architecture

 ```
 backend/
-├── main.py              # FastAPI app with all routes
-├── models.py            # Pydantic request/response models
-├── platform_detect.py   # Platform detection for backend selection
-├── tts.py              # TTS backend abstraction (delegates to MLX or PyTorch)
-├── transcribe.py       # STT backend abstraction (delegates to MLX or PyTorch)
-├── backends/           # Backend implementations
-│   ├── __init__.py     # Backend factory and protocols
-│   ├── mlx_backend.py  # MLX backend (Apple Silicon)
-│   └── pytorch_backend.py  # PyTorch backend (Windows/Linux/Intel)
-├── profiles.py         # Voice profile CRUD
-├── history.py          # Generation history
-├── studio.py           # Audio editing (TODO)
-├── database.py         # SQLite ORM
-└── utils/
-    ├── audio.py        # Audio processing utilities
-    ├── cache.py        # Voice prompt caching
-    └── validation.py   # Input validation
+  app.py                  # FastAPI app factory, CORS, lifecycle events
+  main.py                 # Entry point (imports app, runs uvicorn)
+  config.py               # Data directory paths and configuration
+  models.py               # Pydantic request/response schemas
+  server.py               # Tauri sidecar launcher, parent-pid watchdog
+
+  routes/                 # Thin HTTP handlers — validation, delegation, response formatting
+  services/               # Business logic, CRUD, orchestration
+  backends/               # TTS/STT engine implementations (MLX, PyTorch, etc.)
+  database/               # ORM models, session management, migrations, seed data
+  utils/                  # Shared utilities (audio, effects, caching, progress tracking)
 ```

-### Backend Selection
-
-Voicebox automatically selects the best backend based on platform:
-
- **Apple Silicon (M1/M2/M3)**: Uses MLX backend with native Metal acceleration (4-5x faster)
- **Windows/Linux/Intel Mac**: Uses PyTorch backend (CUDA GPU if available, CPU fallback)
-
-The backend is detected at runtime via `platform_detect.py`. Both backends implement the same interface, so the API remains consistent across platforms.
-
-## API Endpoints
-
-### Health & Info
-
-#### `GET /`
-Root endpoint with version info.
-
-#### `GET /health`
-Health check with model status.
-
-**Response:**
-```json
-{
-  "status": "healthy",
-  "model_loaded": true,
-  "gpu_available": true,
-  "gpu_type": "Metal (Apple Silicon via MLX)",
-  "backend_type": "mlx",
-  "vram_used_mb": null
-}
-```
-
-**Backend Types:**
- `"mlx"` - MLX backend (Apple Silicon with Metal acceleration)
- `"pytorch"` - PyTorch backend (Windows/Linux/Intel Mac)
-
-### Voice Profiles
-
-**Note:** The database is automatically initialized when the server starts. No manual setup required.
-
-#### `POST /profiles`
-Create a new voice profile.
-
-**Request:**
-```json
-{
-  "name": "My Voice",
-  "description": "Optional description",
-  "language": "en"
-}
-```
-
-**Response:**
-```json
-{
-  "id": "uuid",
-  "name": "My Voice",
-  "description": "Optional description",
-  "language": "en",
-  "created_at": "2024-01-01T00:00:00Z",
-  "updated_at": "2024-01-01T00:00:00Z"
-}
-```
-
-#### `GET /profiles`
-List all voice profiles.
-
-#### `GET /profiles/{profile_id}`
-Get a specific profile.
-
-#### `PUT /profiles/{profile_id}`
-Update a profile.
-
-#### `DELETE /profiles/{profile_id}`
-Delete a profile and all associated samples.
-
-#### `POST /profiles/{profile_id}/samples`
-Add a sample to a profile.
-
-**Form Data:**
- `file`: Audio file (WAV, MP3, etc.)
- `reference_text`: Transcript of the audio
-
-**Response:**
-```json
-{
-  "id": "sample-uuid",
-  "profile_id": "profile-uuid",
-  "audio_path": "/path/to/sample.wav",
-  "reference_text": "This is my voice"
-}
-```
-
-#### `GET /profiles/{profile_id}/samples`
-List all samples for a profile.
-
-#### `DELETE /profiles/samples/{sample_id}`
-Delete a specific sample.
-
-### Generation
-
-#### `POST /generate`
-Generate speech from text using a voice profile.
-
-**Request:**
-```json
-{
-  "profile_id": "uuid",
-  "text": "Hello, this is a test.",
-  "language": "en",
-  "seed": 42
-}
-```
-
-**Response:**
-```json
-{
-  "id": "generation-uuid",
-  "profile_id": "profile-uuid",
-  "text": "Hello, this is a test.",
-  "language": "en",
-  "audio_path": "/path/to/audio.wav",
-  "duration": 2.5,
-  "seed": 42,
-  "created_at": "2024-01-01T00:00:00Z"
-}
-```
-
-### History
-
-#### `GET /history`
-List generation history with optional filters.
-
-**Query Parameters:**
- `profile_id` (optional): Filter by profile
- `search` (optional): Search in text content
- `limit` (default: 50): Results per page
- `offset` (default: 0): Pagination offset
-
-#### `GET /history/{generation_id}`
-Get a specific generation.
-
-#### `DELETE /history/{generation_id}`
-Delete a generation.
-
-#### `GET /history/stats`
-Get generation statistics.
-
-**Response:**
-```json
-{
-  "total_generations": 100,
-  "total_duration_seconds": 250.5,
-  "generations_by_profile": {
-    "profile-uuid-1": 50,
-    "profile-uuid-2": 50
-  }
-}
-```
-
-### Audio Files
-
-#### `GET /audio/{generation_id}`
-Download generated audio file.
-
-Returns WAV file with appropriate headers.
-
-### Transcription
-
-#### `POST /transcribe`
-Transcribe audio file to text.
-
-**Form Data:**
- `file`: Audio file
- `language` (optional): Language hint (en or zh)
-
-**Response:**
-```json
-{
-  "text": "Transcribed text here",
-  "duration": 5.5
-}
-```
-
-### Model Management
-
-#### `POST /models/load`
-Manually load TTS model.
-
-**Query Parameters:**
- `model_size`: Model size (1.7B or 0.6B)
-
-#### `POST /models/unload`
-Unload TTS model to free memory.
-
-## Database Schema
-
-### profiles
- `id`: UUID primary key
- `name`: Profile name (unique)
- `description`: Optional description
- `language`: Language code (en/zh)
- `created_at`: Creation timestamp
- `updated_at`: Last update timestamp
-
-### profile_samples
- `id`: UUID primary key
- `profile_id`: Foreign key to profiles
- `audio_path`: Path to audio file
- `reference_text`: Transcript
-
-### generations
- `id`: UUID primary key
- `profile_id`: Foreign key to profiles
- `text`: Generated text
- `language`: Language code
- `audio_path`: Path to audio file
- `duration`: Duration in seconds
- `seed`: Random seed (optional)
- `created_at`: Creation timestamp
-
-### projects
- `id`: UUID primary key
- `name`: Project name
- `data`: JSON data
- `created_at`: Creation timestamp
- `updated_at`: Last update timestamp
-
-## File Structure
+### Request flow

 ```
-data/
-├── profiles/
-│   └── {profile_id}/
-│       ├── {sample_id}.wav
-│       └── ...
-├── generations/
-│   └── {generation_id}.wav
-├── cache/
-│   └── {hash}.prompt
-├── projects/
-│   └── {project_id}.json
-└── voicebox.db
+HTTP request
+  -> routes/        (validate input, parse params)
+  -> services/      (business logic, database queries, orchestration)
+  -> backends/      (TTS/STT inference)
+  -> utils/         (audio processing, effects, caching)
 ```

-## Setup
+Route handlers are intentionally thin. They validate input, delegate to a service function, and format the response. All business logic lives in `services/`.

-### 1. Install Dependencies
-
-```bash
-pip install -r requirements.txt
-```
-
-**Note:** On Apple Silicon, also install MLX dependencies for faster inference:
-```bash
-pip install -r requirements-mlx.txt
-```
-
-### 2. Download Models (Automatic)
-
-The Qwen3-TTS models are automatically downloaded from HuggingFace Hub on first use, similar to how Whisper models work.
-
-**No manual download required!** The models will be cached locally after the first download.
-
-Available models:
- **1.7B** (recommended): `Qwen/Qwen3-TTS-12Hz-1.7B-Base` (~4GB)
- **0.6B** (faster): `Qwen/Qwen3-TTS-12Hz-0.6B-Base` (~2GB)
-
-**Note:** The first generation will take longer as the model downloads. Subsequent generations will use the cached model.
-
-#### Manual Download (Optional)
-
-If you prefer to download models manually or have limited internet during runtime:
-
-```bash
-# Install huggingface-cli
-pip install huggingface_hub
-
-# Download 1.7B model
-huggingface-cli download Qwen/Qwen3-TTS-12Hz-1.7B-Base
-
-# Or use Python
-python -c "from huggingface_hub import snapshot_download; snapshot_download('Qwen/Qwen3-TTS-12Hz-1.7B-Base')"
-```
-
-Models are cached in `~/.cache/huggingface/hub/` by default.
-
-### 4. Run Server
-
-```bash
-# Development (local only)
-python -m backend.main
-
-# Production (allow remote access)
-python -m backend.main --host 0.0.0.0 --port 8000
-```
-
-## Usage Examples
-
-The desktop app, web client, and current development workflow use `http://localhost:17493` by default.
-If you launch the backend manually with a different host or port, substitute that address in the examples below.
-
-### Creating a Voice Profile
-
-```bash
-# 1. Create profile
-curl -X POST http://localhost:17493/profiles \
-  -H "Content-Type: application/json" \
-  -d '{"name": "My Voice", "language": "en"}'
-
-# Response: {"id": "abc-123", ...}
-
-# 2. Add sample
-curl -X POST http://localhost:17493/profiles/abc-123/samples \
-  -F "[email protected]" \
-  -F "reference_text=This is my voice sample"
-```
-
-### Generating Speech
+### Key modules
+
+**services/generation.py** -- Single `run_generation()` function that handles all three generation modes (generate, retry, regenerate). Manages model loading, voice prompt creation, chunked inference, normalization, effects, and version persistence.
+
+**services/task_queue.py** -- Serial generation queue. Ensures only one GPU inference runs at a time. Background tasks are tracked to prevent garbage collection.
+
+**backends/__init__.py** -- Protocol definitions (`TTSBackend`, `STTBackend`), model config registry, and factory functions. Adding a new engine means implementing the protocol and registering a config entry.
+
+**backends/base.py** -- Shared utilities used across all engine implementations: HuggingFace cache checks, device detection, voice prompt combination, progress tracking.
+
+**database/** -- SQLAlchemy ORM models with a re-exporting `__init__.py` for backward compatibility. Migrations run automatically on startup.
+
+### Backend selection
+
+The server detects the best inference backend at startup:
+
+| Platform | Backend | Acceleration |
+|----------|---------|-------------|
+| macOS (Apple Silicon) | MLX | Metal / Neural Engine |
+| Windows / Linux (NVIDIA) | PyTorch | CUDA |
+| Linux (AMD) | PyTorch | ROCm |
+| Intel Arc | PyTorch | IPEX / XPU |
+| Windows (any GPU) | PyTorch | DirectML |
+| Any | PyTorch | CPU fallback |
+
+Detection is handled by `utils/platform_detect.py`. Both backends implement the same `TTSBackend` protocol, so the API layer is engine-agnostic.
+
+## API
+
+90 endpoints organized by domain. Full interactive documentation available at `http://localhost:17493/docs` when the server is running.
+
+| Domain | Prefix | Description |
+|--------|--------|-------------|
+| Health | `/`, `/health` | Server status, GPU info, filesystem checks |
+| Profiles | `/profiles` | Voice profile CRUD, samples, avatars, import/export |
+| Channels | `/channels` | Audio channel management and voice assignment |
+| Generation | `/generate` | TTS generation, retry, regenerate, status SSE |
+| History | `/history` | Generation history, search, favorites, export |
+| Transcription | `/transcribe` | Whisper-based audio-to-text |
+| Stories | `/stories` | Multi-track timeline editor, audio export |
+| Effects | `/effects` | Effect presets, preview, version management |
+| Audio | `/audio`, `/samples` | Audio file serving |
+| Models | `/models` | Load, unload, download, migrate, status |
+| Tasks | `/tasks`, `/cache` | Active task tracking, cache management |
+| CUDA | `/backend/cuda-*` | CUDA binary download and management |
+
+### Quick examples

 ```bash
+# Generate speech
 curl -X POST http://localhost:17493/generate \
  -H "Content-Type: application/json" \
-  -d '{
-    "profile_id": "abc-123",
-    "text": "Hello, this is a test.",
-    "language": "en",
-    "seed": 42
-  }'
+  -d '{"text": "Hello world", "profile_id": "...", "language": "en"}'

-# Response: {"id": "gen-456", "audio_path": "/path/to/audio.wav", ...}
+# List profiles
+curl http://localhost:17493/profiles

-# Download audio
-curl http://localhost:17493/audio/gen-456 -o output.wav
+# Stream generation status (SSE)
+curl http://localhost:17493/generate/{id}/status
 ```

-### Transcribing Audio
+## Data directory
+
+```
+{data_dir}/
+  voicebox.db             # SQLite database
+  profiles/{id}/          # Voice samples per profile
+  generations/            # Generated audio files
+  cache/                  # Voice prompt cache (memory + disk)
+  backends/               # Downloaded CUDA binary (if applicable)
+```
+
+Default location is the OS-specific app data directory. Override with `--data-dir` or the `VOICEBOX_DATA_DIR` environment variable.
+
+## Code quality
+
+Linting and formatting are enforced by [ruff](https://docs.astral.sh/ruff/), configured in `pyproject.toml`. See `STYLE_GUIDE.md` for conventions.

 ```bash
-curl -X POST http://localhost:17493/transcribe \
-  -F "[email protected]" \
-  -F "language=en"
-
-# Response: {"text": "Transcribed text", "duration": 5.5}
+just check-python       # lint + format check
+just fix-python         # auto-fix lint issues + reformat
+just test               # run pytest
 ```

-## Advanced Features
+## Dependencies

-### Multi-Sample Profiles
-
-Add multiple samples to a profile for better quality:
-
-```bash
-# Add first sample
-curl -X POST http://localhost:17493/profiles/abc-123/samples \
-  -F "[email protected]" \
-  -F "reference_text=First sample"
-
-# Add second sample
-curl -X POST http://localhost:17493/profiles/abc-123/samples \
-  -F "[email protected]" \
-  -F "reference_text=Second sample"
-
-# Generation will automatically combine all samples
-```
-
-### Voice Prompt Caching
-
-Voice prompts are automatically cached for faster generation:
- First generation: ~5-10 seconds (creates prompt)
- Subsequent generations: ~1-2 seconds (uses cached prompt)
-
-Cache is stored in `data/cache/` and persists across server restarts.
-
-### VRAM Management
-
-Models are lazy-loaded and can be manually unloaded:
-
-```bash
-# Unload TTS model
-curl -X POST http://localhost:17493/models/unload
-
-# Load specific model size
-curl -X POST "http://localhost:17493/models/load?model_size=0.6B"
-```
-
-## Error Handling
-
-All endpoints return proper HTTP status codes:
-
- `200 OK`: Success
- `400 Bad Request`: Invalid input
- `404 Not Found`: Resource not found
- `500 Internal Server Error`: Server error
-
-Error responses include details:
-
-```json
-{
-  "detail": "Profile not found"
-}
-```
-
-## Performance Tips
-
-1. **Use multi-sample profiles** - Better quality than single sample
-2. **Let caching work** - Voice prompts are cached automatically
-3. **Use 0.6B model on CPU** - Faster than 1.7B with acceptable quality
-4. **Use 1.7B model on GPU** - Best quality, still fast
-5. **Unload Whisper after transcription** - Frees VRAM for TTS
-
-## TODO
-
- [ ] WebSocket support for generation progress
- [ ] Batch generation endpoint
- [ ] Audio effects (M3GAN, etc.)
- [ ] Voice design (text-to-voice)
- [ ] Audio studio timeline features
- [ ] Project management
- [ ] Authentication & rate limiting
- [ ] Export/import profiles
-
-## License
-
-See main project LICENSE.
+Runtime dependencies are in `requirements.txt`. macOS-only MLX dependencies are in `requirements-mlx.txt`. Dev tools (ruff, pytest) are installed automatically by `just setup-python`.
@@ -0,0 +1,404 @@
+# Python Style Guide
+
+Target: **Python 3.12+** | Formatter/Linter: **Ruff** | Config: `backend/pyproject.toml`
+
+This guide codifies the conventions used across the backend, and prescribes the target style for code written during the refactor (Phases 3-6). Existing code should be migrated incrementally -- don't reformat entire files in unrelated PRs.
+
+---
+
+## Formatting
+
+Enforced by `ruff format` (Black-compatible).
+
+- **Line length**: 120 characters.
+- **Indent**: 4 spaces. No tabs.
+- **Trailing commas**: Required on multi-line function signatures, arguments, collections.
+- **Quotes**: Double quotes (`"`) for strings. Single quotes are acceptable in f-string expressions and dict keys inside f-strings where avoiding escapes improves readability.
+
+Run: `ruff format backend/`
+
+---
+
+## Imports
+
+Enforced by ruff's `isort` rules (rule set `I`).
+
+**Grouping** -- three blocks separated by a blank line:
+
+```python
+import asyncio                          # 1. stdlib
+from pathlib import Path
+
+import numpy as np                      # 2. third-party
+from fastapi import APIRouter, HTTPException
+from sqlalchemy.orm import Session
+
+from backend.config import get_data_dir  # 3. local (absolute)
+from .database import get_db            #    or relative
+```
+
+**Rules:**
+- Within the `backend` package, use **relative imports** for sibling/child modules: `from .database import get_db`, `from ..utils.audio import load_audio`.
+- Absolute imports are fine for top-level references from entry points (`main.py`, `server.py`).
+- Never use wildcard imports (`from module import *`).
+- One import per line for `from X import Y` when there are 4+ names; below that, comma-separated is fine.
+- **Lazy imports** are acceptable for heavy dependencies (torch, transformers, mlx) inside functions to reduce startup time. Add a comment: `# lazy: heavy import`.
+
+---
+
+## Type Annotations
+
+Python 3.12 means we use **built-in generics and union syntax natively**. No `from __future__ import annotations`, no `typing.List`/`typing.Dict`.
+
+```python
+# Yes
+def process(items: list[str], config: dict[str, int] | None = None) -> tuple[int, str]: ...
+
+# No
+from typing import List, Dict, Optional, Tuple
+def process(items: List[str], config: Optional[Dict[str, int]] = None) -> Tuple[int, str]: ...
+```
+
+**What to annotate:**
+- All public function signatures (parameters + return type).
+- Private functions: parameters at minimum; return type encouraged.
+- Module-level variables: only when the type isn't obvious from the assignment.
+- Route handlers: parameters are annotated via FastAPI's dependency injection. Add explicit `-> SomeResponse` return types when the route doesn't use `response_model`.
+
+**Imports from `typing` that are still needed** (no built-in equivalent):
+`Literal`, `TypeAlias`, `Protocol`, `runtime_checkable`, `Callable`, `Any`, `ClassVar`, `TypeVar`, `overload`, `TYPE_CHECKING`.
+
+Use `collections.abc` for abstract types: `Sequence`, `Mapping`, `Iterable`, `Iterator`, `Generator`.
+
+---
+
+## Naming
+
+| Thing | Convention | Example |
+|-------|-----------|---------|
+| Module | `snake_case` | `task_queue.py` |
+| Class | `PascalCase` | `ProgressManager` |
+| Function / method | `snake_case` | `create_profile` |
+| Variable | `snake_case` | `sample_rate` |
+| Constant | `UPPER_SNAKE_CASE` | `DEFAULT_SAMPLE_RATE` |
+| Private | `_leading_underscore` | `_generation_queue` |
+| Type alias | `PascalCase` | `EffectChain = list[dict[str, Any]]` |
+
+**Specific conventions:**
+- Database ORM models imported with `DB` prefix alias: `from .database import VoiceProfile as DBVoiceProfile`.
+- Pydantic models use descriptive suffixes: `VoiceProfileCreate`, `VoiceProfileResponse`, `GenerationRequest`.
+- Backend classes use engine-name prefix: `MLXTTSBackend`, `PyTorchSTTBackend`.
+
+---
+
+## Docstrings
+
+**Google style**. Required on all public functions, classes, and modules.
+
+```python
+def combine_voice_prompts(
+    profile_dir: Path,
+    *,
+    target_sr: int = 24000,
+) -> tuple[np.ndarray, int]:
+    """Load and concatenate all voice prompt files for a profile.
+
+    Reads .wav/.mp3/.flac files from the profile directory, resamples to
+    the target sample rate, normalizes, and concatenates into a single array.
+
+    Args:
+        profile_dir: Path to the voice profile directory containing audio files.
+        target_sr: Target sample rate for the output. Defaults to 24000.
+
+    Returns:
+        Tuple of (concatenated audio array, sample rate).
+
+    Raises:
+        FileNotFoundError: If profile_dir does not exist.
+        ValueError: If no valid audio files are found.
+    """
+```
+
+**Short form** is fine for simple functions:
+
+```python
+def get_db_path() -> Path:
+    """Get the path to the SQLite database file."""
+```
+
+**When to skip**: Private helpers under ~5 lines where the name and signature make intent obvious.
+
+**Module docstrings**: A single sentence at the top of every file describing its purpose.
+
+```python
+"""Voice profile CRUD operations."""
+```
+
+---
+
+## Comments
+
+Comments explain **why**, not **what**. If the code needs a comment to explain what it does, the code should be rewritten to be clearer. The exceptions are non-obvious performance choices, external constraints, and concurrency/race-condition reasoning -- those always deserve a comment.
+
+### No section dividers
+
+Do not use ASCII dividers to create visual sections in files:
+
+```python
+# No -- any of these:
+# ============================================
+# GENERATION ENDPOINTS
+# ============================================
+
+# ---------------------------------------------------------------------------
+# Device detection
+# ---------------------------------------------------------------------------
+
+# --- Load model --------------------------------------------------
+```
+
+If a file needs section dividers to be navigable, the file is too long. Split it into modules. Within a function, if you need labeled sections to follow the logic, extract those sections into named functions.
+
+### Inline comments
+
+Inline comments (end-of-line) are fine when they add information the code can't express:
+
+```python
+# Yes -- explains a non-obvious constraint or gives context:
+audio, sr = load_audio(path, sr=24000)  # Qwen expects 24kHz mono
+_generation_queue: asyncio.Queue = None  # type: ignore  # initialized at startup
+"tauri://localhost",         # Tauri webview (macOS)
+
+# No -- restates the code:
+# Check if profile name already exists
+existing = db.query(DBVoiceProfile).filter_by(name=data.name).first()
+
+# Delete from database
+db.delete(sample)
+
+# Update fields
+profile.name = data.name
+```
+
+Delete comments that narrate what the next line of code obviously does. If the function name, variable name, or method call already communicates intent, the comment is noise.
+
+### Block comments
+
+Use block comments for **why** explanations -- constraints, workarounds, non-obvious decisions:
+
+```python
+# PyInstaller + multiprocessing: child processes re-execute the frozen binary
+# with internal arguments. freeze_support() handles this and exits early.
+multiprocessing.freeze_support()
+
+# Mark any stale "generating" records as failed -- these are leftovers
+# from a previous process that was killed mid-generation.
+db.query(Generation).filter_by(status="generating").update({"status": "failed"})
+```
+
+Keep block comments tight. Two to three lines is normal. If you need a paragraph, it probably belongs in the docstring or a design doc.
+
+### Linter/type-checker suppression
+
+Always add a reason after `noqa` and `type: ignore`:
+
+```python
+import intel_extension_for_pytorch  # noqa: F401 -- side-effect import enables XPU
+_queue: asyncio.Queue = None  # type: ignore[assignment]  # initialized at startup
+```
+
+Bare `# noqa` or `# type: ignore` with no explanation are not allowed.
+
+### TODO / FIXME
+
+Use sparingly. Every `TODO` must include a brief description of what needs doing. Don't use them as a substitute for tracking work properly:
+
+```python
+# TODO: replace with async SQLAlchemy once CRUD modules are migrated (Phase 5)
+result = await asyncio.to_thread(profiles.get_profile, profile_id, db)
+```
+
+Never commit `HACK`, `XXX`, or `FIXME` -- fix the problem or file an issue.
+
+### Commented-out code
+
+Delete it. That's what git is for. If you need to document that something was intentionally removed, a short tombstone comment is acceptable:
+
+```python
+# Removed config.json-only check -- too lenient, doesn't confirm weights exist.
+```
+
+---
+
+## Error Handling
+
+The refactor is standardizing on a **two-layer pattern**:
+
+### 1. Domain layer -- raise plain exceptions
+
+CRUD modules and services raise `ValueError`, `FileNotFoundError`, or (post-refactor) custom exceptions defined in `backend/errors.py`:
+
+```python
+# backend/errors.py  (to be created in Phase 4)
+class NotFoundError(Exception):
+    """Raised when a requested resource does not exist."""
+
+class ConflictError(Exception):
+    """Raised on uniqueness constraint violations."""
+```
+
+```python
+# In a service or CRUD module:
+raise NotFoundError(f"Profile {profile_id} not found")
+```
+
+### 2. Route layer -- translate to HTTPException
+
+Route handlers catch domain exceptions and convert:
+
+```python
+@router.post("/profiles")
+async def create_profile(data: VoiceProfileCreate, db: Session = Depends(get_db)):
+    try:
+        return await profiles.create_profile(data, db)
+    except ConflictError as e:
+        raise HTTPException(status_code=409, detail=str(e))
+```
+
+**Background tasks** catch `Exception` broadly, log with `logger.exception()`, and update the task status to `"failed"`.
+
+**Never**: silently swallow exceptions, use bare `except:`, or catch `BaseException`.
+
+---
+
+## Async
+
+### Rules for the refactor
+
+1. **Don't declare `async def` unless the function awaits something.** Several service modules still declare `async def` without awaiting -- these should be migrated to sync functions with `asyncio.to_thread()` at the route layer, or to real async SQLAlchemy.
+2. **CPU-bound work** (audio processing, numpy operations) goes through `asyncio.to_thread()`:
+   ```python
+   audio, sr = await asyncio.to_thread(load_audio, source_path)
+   ```
+3. **GPU-bound TTS inference** is serialized through the generation queue (`services/task_queue.py`). Never call a backend's `generate()` directly from a route handler.
+4. **Fire-and-forget tasks**: use `asyncio.create_task()` and track the task reference to prevent garbage collection:
+   ```python
+   task = asyncio.create_task(some_coro())
+   _background_tasks.add(task)
+   task.add_done_callback(_background_tasks.discard)
+   ```
+
+---
+
+## Logging
+
+Use the `logging` module. Not `print()`.
+
+```python
+import logging
+
+logger = logging.getLogger(__name__)
+
+logger.info("Loading model %s on %s", model_name, device)
+logger.warning("Cache miss for %s, downloading", repo_id)
+logger.exception("Generation %s failed")  # logs traceback automatically
+```
+
+**Rules:**
+- Use `%s`-style placeholders in log calls (not f-strings). This avoids formatting the string if the log level is filtered out.
+- Use `logger.exception()` inside `except` blocks -- it captures the traceback.
+- Logger name should be `__name__` (yields `backend.utils.audio`, etc.).
+- Existing `print()` calls should be migrated to logging as files are touched during the refactor.
+
+---
+
+## Constants
+
+- Define at **module level** in the file where they're primarily used.
+- Use `UPPER_SNAKE_CASE`.
+- Shared/cross-cutting constants (sample rates, file size limits, CORS origins) go in `backend/config.py` after Phase 6 consolidation.
+- Magic numbers in function bodies should be extracted to named constants:
+  ```python
+  # No
+  if len(audio) > 24000 * 60 * 10:
+
+  # Yes
+  MAX_AUDIO_DURATION_SAMPLES = SAMPLE_RATE * 60 * 10
+  if len(audio) > MAX_AUDIO_DURATION_SAMPLES:
+  ```
+
+---
+
+## Function Signatures
+
+- **Keyword-only arguments** (after `*`) for functions with 3+ parameters, especially when several share the same type:
+  ```python
+  def is_model_cached(
+      hf_repo: str,
+      *,
+      weight_extensions: tuple[str, ...] = (".safetensors", ".bin"),
+      required_files: list[str] | None = None,
+  ) -> bool:
+  ```
+- Parameters on **separate lines** when the signature exceeds ~100 characters or has 3+ params.
+- **Trailing comma** after the last parameter in multi-line signatures.
+- Default values inline with the parameter.
+
+---
+
+## String Formatting
+
+- **f-strings** for runtime string construction.
+- **`%s`-style** for `logging` calls (lazy evaluation).
+- **`.format()`**: avoid; f-strings are preferred.
+
+---
+
+## Testing
+
+Framework: **pytest** with `pytest-asyncio`.
+
+- Test files: `test_<module>.py` in `backend/tests/`.
+- Use `conftest.py` for shared fixtures (db sessions, test client, mock backends).
+- Group related tests in classes: `class TestProfileCRUD:`.
+- Use `@pytest.mark.asyncio` for async tests.
+- Use `@pytest.mark.parametrize` to reduce repetition.
+- Manual integration scripts stay in `tests/` but are clearly marked (filename prefix `manual_` or documented in `tests/README.md`).
+
+---
+
+## Project Layout
+
+```
+backend/
+  app.py              # FastAPI app factory, CORS, lifecycle events
+  main.py             # Entry point (imports app, runs uvicorn)
+  config.py           # Data directory paths
+  models.py           # Pydantic request/response schemas
+  server.py           # Tauri sidecar launcher, parent-pid watchdog
+  routes/             # Thin HTTP handlers (validation, delegation, response formatting)
+  services/           # Business logic, CRUD, orchestration
+  backends/           # TTS/STT engine implementations
+  database/           # ORM models, session management, migrations, seeds
+  utils/              # Shared utilities (audio, effects, caching, progress)
+  tests/              # pytest suite
+```
+
+---
+
+## Ruff Adoption
+
+`pyproject.toml` configures ruff for linting and formatting. Run:
+
+```bash
+# Lint (check)
+ruff check backend/
+
+# Lint (auto-fix)
+ruff check backend/ --fix
+
+# Format
+ruff format backend/
+```
+
+Introduce ruff fixes file-by-file as you touch them. Don't run `--fix` across the entire codebase in one shot -- that creates unreviewable diffs.
@@ -1,3 +1,3 @@
 # Backend package

-__version__ = "0.2.1"
+__version__ = "0.2.3"
@@ -0,0 +1,215 @@
+"""FastAPI application factory, middleware, and lifecycle events."""
+
+import asyncio
+import logging
+import os
+import sys
+from pathlib import Path
+
+
+class ColoredFormatter(logging.Formatter):
+    """Custom formatter to add colors matching uvicorn's style."""
+
+    COLORS = {
+        "DEBUG": "\033[36m",  # Cyan
+        "INFO": "\033[32m",  # Green
+        "WARNING": "\033[33m",  # Yellow
+        "ERROR": "\033[31m",  # Red
+        "CRITICAL": "\033[35m",  # Magenta
+    }
+    RESET = "\033[0m"
+
+    def format(self, record):
+        log_color = self.COLORS.get(record.levelname, self.RESET)
+        record.levelname = f"{log_color}{record.levelname}{self.RESET}"
+        return super().format(record)
+
+
+# Configure logging to match uvicorn's format with colors
+handler = logging.StreamHandler(sys.stderr)
+handler.setFormatter(ColoredFormatter("%(levelname)s:     %(message)s"))
+logging.basicConfig(
+    level=logging.INFO,
+    handlers=[handler],
+)
+
+logger = logging.getLogger(__name__)
+
+# AMD GPU environment variables must be set before torch import
+if not os.environ.get("HSA_OVERRIDE_GFX_VERSION"):
+    os.environ["HSA_OVERRIDE_GFX_VERSION"] = "10.3.0"
+if not os.environ.get("MIOPEN_LOG_LEVEL"):
+    os.environ["MIOPEN_LOG_LEVEL"] = "4"
+
+import torch
+from fastapi import FastAPI
+from fastapi.middleware.cors import CORSMiddleware
+from urllib.parse import quote
+
+from . import __version__, config, database
+from .services import tts, transcribe
+from .database import get_db
+from .utils.platform_detect import get_backend_type
+from .utils.progress import get_progress_manager
+from .services.task_queue import create_background_task, init_queue
+from .routes import register_routers
+
+
+def safe_content_disposition(disposition_type: str, filename: str) -> str:
+    """Build a Content-Disposition header safe for non-ASCII filenames.
+
+    Uses RFC 5987 ``filename*`` parameter so browsers can decode UTF-8
+    filenames while the ``filename`` fallback stays ASCII-only.
+    """
+    ascii_name = "".join(c for c in filename if c.isascii() and (c.isalnum() or c in " -_.")).strip() or "download"
+    utf8_name = quote(filename, safe="")
+    return f"{disposition_type}; filename=\"{ascii_name}\"; filename*=UTF-8''{utf8_name}"
+
+
+def create_app() -> FastAPI:
+    """Create and configure the FastAPI application."""
+    application = FastAPI(
+        title="voicebox API",
+        description="Production-quality Qwen3-TTS voice cloning API",
+        version=__version__,
+    )
+
+    _configure_cors(application)
+    register_routers(application)
+    _register_lifecycle(application)
+
+    return application
+
+
+def _configure_cors(application: FastAPI) -> None:
+    """Set up CORS middleware with local-first defaults."""
+    default_origins = [
+        "http://localhost:5173",  # Vite dev server
+        "http://127.0.0.1:5173",
+        "http://localhost:17493",
+        "http://127.0.0.1:17493",
+        "tauri://localhost",  # Tauri webview (macOS)
+        "https://tauri.localhost",  # Tauri webview (Windows/Linux)
+        "http://tauri.localhost",  # Tauri webview (Windows, some builds)
+    ]
+    env_origins = os.environ.get("VOICEBOX_CORS_ORIGINS", "")
+    all_origins = default_origins + [o.strip() for o in env_origins.split(",") if o.strip()]
+
+    application.add_middleware(
+        CORSMiddleware,
+        allow_origins=all_origins,
+        allow_credentials=True,
+        allow_methods=["*"],
+        allow_headers=["*"],
+    )
+
+
+def _get_gpu_status() -> str:
+    """Return a human-readable string describing GPU availability."""
+    backend_type = get_backend_type()
+    if torch.cuda.is_available():
+        device_name = torch.cuda.get_device_name(0)
+        is_rocm = hasattr(torch.version, "hip") and torch.version.hip is not None
+        if is_rocm:
+            return f"ROCm ({device_name})"
+        return f"CUDA ({device_name})"
+    elif hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
+        return "MPS (Apple Silicon)"
+    elif backend_type == "mlx":
+        return "Metal (Apple Silicon via MLX)"
+    return "None (CPU only)"
+
+
+def _register_lifecycle(application: FastAPI) -> None:
+    """Attach startup and shutdown event handlers."""
+
+    @application.on_event("startup")
+    async def startup_event():
+        import platform
+        import sys
+
+        logger.info("Voicebox v%s starting up", __version__)
+        logger.info(
+            "Python %s on %s %s (%s)",
+            sys.version.split()[0],
+            platform.system(),
+            platform.release(),
+            platform.machine(),
+        )
+
+        database.init_db()
+
+        from .database.session import _db_path
+
+        logger.info("Database: %s", _db_path)
+        logger.info("Data directory: %s", config.get_data_dir())
+
+        init_queue()
+
+        # Mark stale "generating" records as failed -- leftovers from a killed process
+        from sqlalchemy import text as sa_text
+
+        db = next(get_db())
+        try:
+            result = db.execute(
+                sa_text(
+                    "UPDATE generations SET status = 'failed', "
+                    "error = 'Server was shut down during generation' "
+                    "WHERE status IN ('generating', 'loading_model')"
+                )
+            )
+            if result.rowcount > 0:
+                logger.info("Marked %d stale generation(s) as failed", result.rowcount)
+
+            from .database import VoiceProfile as DBVoiceProfile, Generation as DBGeneration
+
+            profile_count = db.query(DBVoiceProfile).count()
+            generation_count = db.query(DBGeneration).count()
+            logger.info("Profiles: %d, Generations: %d", profile_count, generation_count)
+
+            db.commit()
+        except Exception as e:
+            db.rollback()
+            logger.warning("Could not clean up stale generations: %s", e)
+        finally:
+            db.close()
+
+        backend_type = get_backend_type()
+        logger.info("Backend: %s", backend_type.upper())
+        logger.info("GPU: %s", _get_gpu_status())
+
+        from .services.cuda import check_and_update_cuda_binary
+
+        create_background_task(check_and_update_cuda_binary())
+
+        try:
+            progress_manager = get_progress_manager()
+            progress_manager._set_main_loop(asyncio.get_running_loop())
+        except Exception as e:
+            logger.warning("Could not initialize progress manager event loop: %s", e)
+
+        try:
+            from huggingface_hub import constants as hf_constants
+
+            cache_dir = Path(hf_constants.HF_HUB_CACHE)
+            cache_dir.mkdir(parents=True, exist_ok=True)
+            logger.info("Model cache: %s", cache_dir)
+        except Exception as e:
+            logger.warning("Could not create HuggingFace cache directory: %s", e)
+
+        logger.info("Ready")
+
+    @application.on_event("shutdown")
+    async def shutdown_event():
+        logger.info("Voicebox server shutting down...")
+        try:
+            tts.unload_tts_model()
+        except Exception:
+            logger.exception("Failed to unload TTS model")
+        try:
+            transcribe.unload_whisper_model()
+        except Exception:
+            logger.exception("Failed to unload Whisper model")
+
+
+app = create_app()
@@ -1,25 +1,66 @@
 """
 Backend abstraction layer for TTS and STT.

-Provides a unified interface for MLX and PyTorch backends.
+Provides a unified interface for MLX and PyTorch backends,
+and a model config registry that eliminates per-engine dispatch maps.
 """

 import threading
+from dataclasses import dataclass, field
 from typing import Protocol, Optional, Tuple, List
 from typing_extensions import runtime_checkable
 import numpy as np

-from ..platform_detect import get_backend_type
+from ..utils.platform_detect import get_backend_type
+
+LANGUAGE_CODE_TO_NAME = {
+    "zh": "chinese",
+    "en": "english",
+    "ja": "japanese",
+    "ko": "korean",
+    "de": "german",
+    "fr": "french",
+    "ru": "russian",
+    "pt": "portuguese",
+    "es": "spanish",
+    "it": "italian",
+}
+
+WHISPER_HF_REPOS = {
+    "base": "openai/whisper-base",
+    "small": "openai/whisper-small",
+    "medium": "openai/whisper-medium",
+    "large": "openai/whisper-large-v3",
+    "turbo": "openai/whisper-large-v3-turbo",
+}
+
+
+@dataclass
+class ModelConfig:
+    """Declarative config for a downloadable model variant."""
+
+    model_name: str  # e.g. "luxtts", "chatterbox-tts"
+    display_name: str  # e.g. "LuxTTS (Fast, CPU-friendly)"
+    engine: str  # e.g. "luxtts", "chatterbox"
+    hf_repo_id: str  # e.g. "YatharthS/LuxTTS"
+    model_size: str = "default"
+    size_mb: int = 0
+    needs_trim: bool = False
+    supports_instruct: bool = False
+    languages: list[str] = field(default_factory=lambda: ["en"])


@runtime_checkable
 class TTSBackend(Protocol):
    """Protocol for TTS backend implementations."""
-    
+
+    # Each backend class should define MODEL_CONFIGS as a class variable:
+    # MODEL_CONFIGS: list[ModelConfig]
+
    async def load_model(self, model_size: str) -> None:
        """Load TTS model."""
        ...
-    
+
    async def create_voice_prompt(
        self,
        audio_path: str,
@@ -28,12 +69,12 @@ class TTSBackend(Protocol):
    ) -> Tuple[dict, bool]:
        """
        Create voice prompt from reference audio.
-        
+
        Returns:
            Tuple of (voice_prompt_dict, was_cached)
        """
        ...
-    
+
    async def combine_voice_prompts(
        self,
        audio_paths: List[str],
@@ -41,12 +82,12 @@ class TTSBackend(Protocol):
    ) -> Tuple[np.ndarray, str]:
        """
        Combine multiple voice prompts.
-        
+
        Returns:
            Tuple of (combined_audio_array, combined_text)
        """
        ...
-    
+
    async def generate(
        self,
        text: str,
@@ -57,24 +98,24 @@ class TTSBackend(Protocol):
    ) -> Tuple[np.ndarray, int]:
        """
        Generate audio from text.
-        
+
        Returns:
            Tuple of (audio_array, sample_rate)
        """
        ...
-    
+
    def unload_model(self) -> None:
        """Unload model to free memory."""
        ...
-    
+
    def is_loaded(self) -> bool:
        """Check if model is loaded."""
        ...
-    
+
    def _get_model_path(self, model_size: str) -> str:
        """
        Get model path for a given size.
-        
+
        Returns:
            Model path or HuggingFace Hub ID
        """
@@ -84,11 +125,11 @@ class TTSBackend(Protocol):
@runtime_checkable
 class STTBackend(Protocol):
    """Protocol for STT (Speech-to-Text) backend implementations."""
-    
+
    async def load_model(self, model_size: str) -> None:
        """Load STT model."""
        ...
-    
+
    async def transcribe(
        self,
        audio_path: str,
@@ -96,16 +137,16 @@ class STTBackend(Protocol):
    ) -> str:
        """
        Transcribe audio to text.
-        
+
        Returns:
            Transcribed text
        """
        ...
-    
+
    def unload_model(self) -> None:
        """Unload model to free memory."""
        ...
-    
+
    def is_loaded(self) -> bool:
        """Check if model is loaded."""
        ...
@@ -117,7 +158,8 @@ _tts_backends: dict[str, TTSBackend] = {}
 _tts_backends_lock = threading.Lock()
 _stt_backend: Optional[STTBackend] = None

-# Supported TTS engines
+# Supported TTS engines — keyed by engine name, value is the backend class import path.
+# The factory function uses this for the if/elif chain; the model configs live on the backend classes.
 TTS_ENGINES = {
    "qwen": "Qwen TTS",
    "luxtts": "LuxTTS",
@@ -126,10 +168,277 @@ TTS_ENGINES = {
 }


+def _get_qwen_model_configs() -> list[ModelConfig]:
+    """Return Qwen model configs with backend-aware HF repo IDs."""
+    backend_type = get_backend_type()
+    if backend_type == "mlx":
+        repo_1_7b = "mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16"
+        repo_0_6b = "mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16"  # 0.6B not available in MLX, falls back
+    else:
+        repo_1_7b = "Qwen/Qwen3-TTS-12Hz-1.7B-Base"
+        repo_0_6b = "Qwen/Qwen3-TTS-12Hz-0.6B-Base"
+
+    return [
+        ModelConfig(
+            model_name="qwen-tts-1.7B",
+            display_name="Qwen TTS 1.7B",
+            engine="qwen",
+            hf_repo_id=repo_1_7b,
+            model_size="1.7B",
+            size_mb=3500,
+            supports_instruct=False,  # Base model drops instruct silently
+            languages=["zh", "en", "ja", "ko", "de", "fr", "ru", "pt", "es", "it"],
+        ),
+        ModelConfig(
+            model_name="qwen-tts-0.6B",
+            display_name="Qwen TTS 0.6B",
+            engine="qwen",
+            hf_repo_id=repo_0_6b,
+            model_size="0.6B",
+            size_mb=1200,
+            supports_instruct=False,
+            languages=["zh", "en", "ja", "ko", "de", "fr", "ru", "pt", "es", "it"],
+        ),
+    ]
+
+
+def _get_non_qwen_tts_configs() -> list[ModelConfig]:
+    """Return model configs for non-Qwen TTS engines.
+
+    These are static — no backend-type branching needed.
+    """
+    return [
+        ModelConfig(
+            model_name="luxtts",
+            display_name="LuxTTS (Fast, CPU-friendly)",
+            engine="luxtts",
+            hf_repo_id="YatharthS/LuxTTS",
+            size_mb=300,
+            languages=["en"],
+        ),
+        ModelConfig(
+            model_name="chatterbox-tts",
+            display_name="Chatterbox TTS (Multilingual)",
+            engine="chatterbox",
+            hf_repo_id="ResembleAI/chatterbox",
+            size_mb=3200,
+            needs_trim=True,
+            languages=[
+                "zh",
+                "en",
+                "ja",
+                "ko",
+                "de",
+                "fr",
+                "ru",
+                "pt",
+                "es",
+                "it",
+                "he",
+                "ar",
+                "da",
+                "el",
+                "fi",
+                "hi",
+                "ms",
+                "nl",
+                "no",
+                "pl",
+                "sv",
+                "sw",
+                "tr",
+            ],
+        ),
+        ModelConfig(
+            model_name="chatterbox-turbo",
+            display_name="Chatterbox Turbo (English, Tags)",
+            engine="chatterbox_turbo",
+            hf_repo_id="ResembleAI/chatterbox-turbo",
+            size_mb=1500,
+            needs_trim=True,
+            languages=["en"],
+        ),
+    ]
+
+
+def _get_whisper_configs() -> list[ModelConfig]:
+    """Return Whisper STT model configs."""
+    return [
+        ModelConfig(
+            model_name="whisper-base",
+            display_name="Whisper Base",
+            engine="whisper",
+            hf_repo_id="openai/whisper-base",
+            model_size="base",
+        ),
+        ModelConfig(
+            model_name="whisper-small",
+            display_name="Whisper Small",
+            engine="whisper",
+            hf_repo_id="openai/whisper-small",
+            model_size="small",
+        ),
+        ModelConfig(
+            model_name="whisper-medium",
+            display_name="Whisper Medium",
+            engine="whisper",
+            hf_repo_id="openai/whisper-medium",
+            model_size="medium",
+        ),
+        ModelConfig(
+            model_name="whisper-large",
+            display_name="Whisper Large",
+            engine="whisper",
+            hf_repo_id="openai/whisper-large-v3",
+            model_size="large",
+        ),
+        ModelConfig(
+            model_name="whisper-turbo",
+            display_name="Whisper Turbo",
+            engine="whisper",
+            hf_repo_id="openai/whisper-large-v3-turbo",
+            model_size="turbo",
+        ),
+    ]
+
+
+def get_all_model_configs() -> list[ModelConfig]:
+    """Return the full list of model configs (TTS + STT)."""
+    return _get_qwen_model_configs() + _get_non_qwen_tts_configs() + _get_whisper_configs()
+
+
+def get_tts_model_configs() -> list[ModelConfig]:
+    """Return only TTS model configs."""
+    return _get_qwen_model_configs() + _get_non_qwen_tts_configs()
+
+
+# Lookup helpers — these replace the if/elif chains in main.py
+
+
+def get_model_config(model_name: str) -> Optional[ModelConfig]:
+    """Look up a model config by model_name."""
+    for cfg in get_all_model_configs():
+        if cfg.model_name == model_name:
+            return cfg
+    return None
+
+
+def engine_needs_trim(engine: str) -> bool:
+    """Whether this engine's output should be run through trim_tts_output."""
+    for cfg in get_tts_model_configs():
+        if cfg.engine == engine:
+            return cfg.needs_trim
+    return False
+
+
+def engine_has_model_sizes(engine: str) -> bool:
+    """Whether this engine supports multiple model sizes (only Qwen currently)."""
+    configs = [c for c in get_tts_model_configs() if c.engine == engine]
+    return len(configs) > 1
+
+
+async def load_engine_model(engine: str, model_size: str = "default") -> None:
+    """Load a model for the given engine, handling the Qwen model_size special case."""
+    backend = get_tts_backend_for_engine(engine)
+    if engine == "qwen":
+        await backend.load_model_async(model_size)
+    else:
+        await backend.load_model()
+
+
+async def ensure_model_cached_or_raise(engine: str, model_size: str = "default") -> None:
+    """Check if a model is cached, raise HTTPException if not. Used by streaming endpoint."""
+    from fastapi import HTTPException
+
+    backend = get_tts_backend_for_engine(engine)
+    cfg = None
+    for c in get_tts_model_configs():
+        if c.engine == engine and c.model_size == model_size:
+            cfg = c
+            break
+
+    if engine == "qwen":
+        if not backend._is_model_cached(model_size):
+            raise HTTPException(
+                status_code=400,
+                detail=f"Model {model_size} is not downloaded yet. Use /generate to trigger a download.",
+            )
+    else:
+        if not backend._is_model_cached():
+            display = cfg.display_name if cfg else engine
+            raise HTTPException(
+                status_code=400,
+                detail=f"{display} model is not downloaded yet. Use /generate to trigger a download.",
+            )
+
+
+def unload_model_by_config(config: ModelConfig) -> bool:
+    """Unload a model given its config. Returns True if it was loaded, False otherwise."""
+    from . import get_tts_backend_for_engine
+    from ..services import tts, transcribe
+
+    if config.engine == "whisper":
+        whisper_model = transcribe.get_whisper_model()
+        if whisper_model.is_loaded() and whisper_model.model_size == config.model_size:
+            transcribe.unload_whisper_model()
+            return True
+        return False
+
+    if config.engine == "qwen":
+        tts_model = tts.get_tts_model()
+        loaded_size = getattr(tts_model, "_current_model_size", None) or getattr(tts_model, "model_size", None)
+        if tts_model.is_loaded() and loaded_size == config.model_size:
+            tts.unload_tts_model()
+            return True
+        return False
+
+    # All other TTS engines
+    backend = get_tts_backend_for_engine(config.engine)
+    if backend.is_loaded():
+        backend.unload_model()
+        return True
+    return False
+
+
+def check_model_loaded(config: ModelConfig) -> bool:
+    """Check if a model is currently loaded."""
+    from . import get_tts_backend_for_engine
+    from ..services import tts, transcribe
+
+    try:
+        if config.engine == "whisper":
+            whisper_model = transcribe.get_whisper_model()
+            return whisper_model.is_loaded() and getattr(whisper_model, "model_size", None) == config.model_size
+
+        if config.engine == "qwen":
+            tts_model = tts.get_tts_model()
+            loaded_size = getattr(tts_model, "_current_model_size", None) or getattr(tts_model, "model_size", None)
+            return tts_model.is_loaded() and loaded_size == config.model_size
+
+        backend = get_tts_backend_for_engine(config.engine)
+        return backend.is_loaded()
+    except Exception:
+        return False
+
+
+def get_model_load_func(config: ModelConfig):
+    """Return a callable that loads/downloads the model."""
+    from . import get_tts_backend_for_engine
+    from ..services import tts, transcribe
+
+    if config.engine == "whisper":
+        return lambda: transcribe.get_whisper_model().load_model(config.model_size)
+
+    if config.engine == "qwen":
+        return lambda: tts.get_tts_model().load_model(config.model_size)
+
+    return lambda: get_tts_backend_for_engine(config.engine).load_model()
+
+
 def get_tts_backend() -> TTSBackend:
    """
    Get or create the default (Qwen) TTS backend instance based on platform.
-    
+
    Returns:
        TTS backend instance (MLX or PyTorch)
    """
@@ -139,45 +448,50 @@ def get_tts_backend() -> TTSBackend:
 def get_tts_backend_for_engine(engine: str) -> TTSBackend:
    """
    Get or create a TTS backend for the given engine.
-    
+
    Args:
-        engine: Engine name ("qwen" or "luxtts")
-    
+        engine: Engine name (e.g. "qwen", "luxtts", "chatterbox", "chatterbox_turbo")
+
    Returns:
        TTS backend instance
    """
    global _tts_backends
-    
+
    # Fast path: check without lock
    if engine in _tts_backends:
        return _tts_backends[engine]
-    
+
    # Slow path: create with lock to avoid duplicate instantiation
    with _tts_backends_lock:
        # Double-check after acquiring lock
        if engine in _tts_backends:
            return _tts_backends[engine]
-        
+
        if engine == "qwen":
            backend_type = get_backend_type()
            if backend_type == "mlx":
                from .mlx_backend import MLXTTSBackend
+
                backend = MLXTTSBackend()
            else:
                from .pytorch_backend import PyTorchTTSBackend
+
                backend = PyTorchTTSBackend()
        elif engine == "luxtts":
            from .luxtts_backend import LuxTTSBackend
+
            backend = LuxTTSBackend()
        elif engine == "chatterbox":
            from .chatterbox_backend import ChatterboxTTSBackend
+
            backend = ChatterboxTTSBackend()
        elif engine == "chatterbox_turbo":
            from .chatterbox_turbo_backend import ChatterboxTurboTTSBackend
+
            backend = ChatterboxTurboTTSBackend()
        else:
            raise ValueError(f"Unknown TTS engine: {engine}. Supported: {list(TTS_ENGINES.keys())}")
-        
+
        _tts_backends[engine] = backend
        return backend

@@ -185,22 +499,24 @@ def get_tts_backend_for_engine(engine: str) -> TTSBackend:
 def get_stt_backend() -> STTBackend:
    """
    Get or create STT backend instance based on platform.
-    
+
    Returns:
        STT backend instance (MLX or PyTorch)
    """
    global _stt_backend
-    
+
    if _stt_backend is None:
        backend_type = get_backend_type()
-        
+
        if backend_type == "mlx":
            from .mlx_backend import MLXSTTBackend
+
            _stt_backend = MLXSTTBackend()
        else:
            from .pytorch_backend import PyTorchSTTBackend
+
            _stt_backend = PyTorchSTTBackend()
-    
+
    return _stt_backend


@@ -0,0 +1,258 @@
+"""
+Shared utilities for TTS/STT backend implementations.
+
+Eliminates duplication of cache checking, device detection,
+voice prompt combination, and model loading progress tracking.
+"""
+
+import logging
+import platform
+from contextlib import contextmanager
+from pathlib import Path
+from typing import Callable, List, Optional, Tuple
+
+import numpy as np
+
+from ..utils.audio import normalize_audio, load_audio
+from ..utils.progress import get_progress_manager
+from ..utils.hf_progress import HFProgressTracker, create_hf_progress_callback
+from ..utils.tasks import get_task_manager
+
+logger = logging.getLogger(__name__)
+
+
+def is_model_cached(
+    hf_repo: str,
+    *,
+    weight_extensions: tuple[str, ...] = (".safetensors", ".bin"),
+    required_files: Optional[list[str]] = None,
+) -> bool:
+    """
+    Check if a HuggingFace model is fully cached locally.
+
+    Args:
+        hf_repo: HuggingFace repo ID (e.g. "Qwen/Qwen3-TTS-12Hz-1.7B-Base")
+        weight_extensions: File extensions that count as model weights.
+        required_files: If set, check that these specific filenames exist
+                        in snapshots instead of checking by extension.
+
+    Returns:
+        True if model is fully cached, False if missing or incomplete.
+    """
+    try:
+        from huggingface_hub import constants as hf_constants
+
+        repo_cache = Path(hf_constants.HF_HUB_CACHE) / ("models--" + hf_repo.replace("/", "--"))
+
+        if not repo_cache.exists():
+            return False
+
+        # Incomplete blobs mean a download is still in progress
+        blobs_dir = repo_cache / "blobs"
+        if blobs_dir.exists() and any(blobs_dir.glob("*.incomplete")):
+            logger.debug(f"Found .incomplete files for {hf_repo}")
+            return False
+
+        snapshots_dir = repo_cache / "snapshots"
+        if not snapshots_dir.exists():
+            return False
+
+        if required_files:
+            # Check that every required filename exists somewhere in snapshots
+            for fname in required_files:
+                if not any(snapshots_dir.rglob(fname)):
+                    return False
+            return True
+
+        # Check that at least one weight file exists
+        for ext in weight_extensions:
+            if any(snapshots_dir.rglob(f"*{ext}")):
+                return True
+
+        logger.debug(f"No model weights found for {hf_repo}")
+        return False
+
+    except Exception as e:
+        logger.warning(f"Error checking cache for {hf_repo}: {e}")
+        return False
+
+
+def get_torch_device(
+    *,
+    allow_xpu: bool = False,
+    allow_directml: bool = False,
+    allow_mps: bool = False,
+    force_cpu_on_mac: bool = False,
+) -> str:
+    """
+    Detect the best available torch device.
+
+    Args:
+        allow_xpu: Check for Intel XPU (IPEX) support.
+        allow_directml: Check for DirectML (Windows) support.
+        allow_mps: Allow MPS (Apple Silicon). If False, MPS falls back to CPU.
+        force_cpu_on_mac: Force CPU on macOS regardless of GPU availability.
+    """
+    if force_cpu_on_mac and platform.system() == "Darwin":
+        return "cpu"
+
+    import torch
+
+    if torch.cuda.is_available():
+        return "cuda"
+
+    if allow_xpu:
+        try:
+            import intel_extension_for_pytorch  # noqa: F401
+
+            if hasattr(torch, "xpu") and torch.xpu.is_available():
+                return "xpu"
+        except ImportError:
+            pass
+
+    if allow_directml:
+        try:
+            import torch_directml
+
+            if torch_directml.device_count() > 0:
+                return torch_directml.device(0)
+        except ImportError:
+            pass
+
+    if allow_mps:
+        if hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
+            return "mps"
+
+    return "cpu"
+
+
+async def combine_voice_prompts(
+    audio_paths: List[str],
+    reference_texts: List[str],
+    *,
+    sample_rate: Optional[int] = None,
+) -> Tuple[np.ndarray, str]:
+    """
+    Combine multiple reference audio samples into one.
+
+    Loads each audio file, normalizes, concatenates, and joins texts.
+
+    Args:
+        audio_paths: Paths to reference audio files.
+        reference_texts: Corresponding transcripts.
+        sample_rate: If set, resample audio to this rate during loading.
+    """
+    combined_audio = []
+
+    for path in audio_paths:
+        kwargs = {"sample_rate": sample_rate} if sample_rate else {}
+        audio, _sr = load_audio(path, **kwargs)
+        audio = normalize_audio(audio)
+        combined_audio.append(audio)
+
+    mixed = np.concatenate(combined_audio)
+    mixed = normalize_audio(mixed)
+    combined_text = " ".join(reference_texts)
+
+    return mixed, combined_text
+
+
+@contextmanager
+def model_load_progress(
+    model_name: str,
+    is_cached: bool,
+    filter_non_downloads: Optional[bool] = None,
+):
+    """
+    Context manager for model loading with HF download progress tracking.
+
+    Handles the tqdm patching, progress_manager/task_manager lifecycle,
+    and error reporting that every backend duplicates.
+
+    Args:
+        model_name: Progress tracking key (e.g. "qwen-tts-1.7B", "whisper-base").
+        is_cached: Whether the model is already downloaded.
+        filter_non_downloads: Whether to filter non-download tqdm bars.
+                              Defaults to `is_cached`.
+
+    Yields:
+        The tracker context (already entered). The caller loads the model
+        inside the `with` block. The tqdm patch is torn down on exit.
+
+    Usage:
+        with model_load_progress("qwen-tts-1.7B", is_cached) as ctx:
+            self.model = SomeModel.from_pretrained(...)
+    """
+    if filter_non_downloads is None:
+        filter_non_downloads = is_cached
+
+    progress_manager = get_progress_manager()
+    task_manager = get_task_manager()
+
+    progress_callback = create_hf_progress_callback(model_name, progress_manager)
+    tracker = HFProgressTracker(progress_callback, filter_non_downloads=filter_non_downloads)
+
+    tracker_context = tracker.patch_download()
+    tracker_context.__enter__()
+
+    if not is_cached:
+        task_manager.start_download(model_name)
+        progress_manager.update_progress(
+            model_name=model_name,
+            current=0,
+            total=0,
+            filename="Connecting to HuggingFace...",
+            status="downloading",
+        )
+
+    try:
+        yield tracker_context
+    except Exception as e:
+        # Report error to both managers
+        progress_manager.mark_error(model_name, str(e))
+        task_manager.error_download(model_name, str(e))
+        raise
+    else:
+        # Only mark complete if we were tracking a download
+        if not is_cached:
+            progress_manager.mark_complete(model_name)
+            task_manager.complete_download(model_name)
+    finally:
+        tracker_context.__exit__(None, None, None)
+
+
+def patch_chatterbox_f32(model) -> None:
+    """
+    Patch float64 -> float32 dtype mismatches in upstream chatterbox.
+
+    librosa.load returns float64 numpy arrays. Multiple upstream code paths
+    convert these to torch tensors via torch.from_numpy() without casting,
+    then matmul against float32 model weights. This patches the two known
+    entry points:
+
+    1. S3Tokenizer.log_mel_spectrogram — audio tensor hits _mel_filters (f32)
+    2. VoiceEncoder.forward — float64 mel spectrograms hit LSTM weights (f32)
+    """
+    import types
+
+    # Patch S3Tokenizer
+    _tokzr = model.s3gen.tokenizer
+    _orig_log_mel = _tokzr.log_mel_spectrogram.__func__
+
+    def _f32_log_mel(self_tokzr, audio, padding=0):
+        import torch as _torch
+
+        if _torch.is_tensor(audio):
+            audio = audio.float()
+        return _orig_log_mel(self_tokzr, audio, padding)
+
+    _tokzr.log_mel_spectrogram = types.MethodType(_f32_log_mel, _tokzr)
+
+    # Patch VoiceEncoder
+    _ve = model.ve
+    _orig_ve_forward = _ve.forward.__func__
+
+    def _f32_ve_forward(self_ve, mels):
+        return _orig_ve_forward(self_ve, mels.float())
+
+    _ve.forward = types.MethodType(_f32_ve_forward, _ve)
@@ -8,7 +8,6 @@ on macOS due to known MPS tensor issues.

 import asyncio
 import logging
-import platform
 import threading
 from pathlib import Path
 from typing import ClassVar, List, Optional, Tuple
@@ -16,9 +15,13 @@ from typing import ClassVar, List, Optional, Tuple
 import numpy as np

 from . import TTSBackend
-from ..utils.audio import normalize_audio, load_audio
-from ..utils.progress import get_progress_manager
-from ..utils.tasks import get_task_manager
+from .base import (
+    is_model_cached,
+    get_torch_device,
+    combine_voice_prompts as _combine_voice_prompts,
+    model_load_progress,
+    patch_chatterbox_f32,
+)

 logger = logging.getLogger(__name__)

@@ -45,17 +48,7 @@ class ChatterboxTTSBackend:
        self._model_load_lock = asyncio.Lock()

    def _get_device(self) -> str:
-        """Get the best available device. Forces CPU on macOS (MPS issue)."""
-        if platform.system() == "Darwin":
-            return "cpu"
-        try:
-            import torch
-
-            if torch.cuda.is_available():
-                return "cuda"
-        except ImportError:
-            pass
-        return "cpu"
+        return get_torch_device(force_cpu_on_mac=True)

    def is_loaded(self) -> bool:
        return self.model is not None
@@ -64,33 +57,7 @@ class ChatterboxTTSBackend:
        return CHATTERBOX_HF_REPO

    def _is_model_cached(self, model_size: str = "default") -> bool:
-        """Check if the Chatterbox multilingual model is cached locally."""
-        try:
-            from huggingface_hub import constants as hf_constants
-
-            repo_cache = Path(hf_constants.HF_HUB_CACHE) / (
-                "models--" + CHATTERBOX_HF_REPO.replace("/", "--")
-            )
-
-            if not repo_cache.exists():
-                return False
-
-            blobs_dir = repo_cache / "blobs"
-            if blobs_dir.exists() and any(blobs_dir.glob("*.incomplete")):
-                return False
-
-            # Check for multilingual weight files
-            snapshots_dir = repo_cache / "snapshots"
-            if snapshots_dir.exists():
-                for fname in _MTL_WEIGHT_FILES:
-                    if not any(snapshots_dir.rglob(fname)):
-                        return False
-                return True
-
-            return False
-        except Exception as e:
-            logger.warning(f"Error checking Chatterbox cache: {e}")
-            return False
+        return is_model_cached(CHATTERBOX_HF_REPO, required_files=_MTL_WEIGHT_FILES)

    async def load_model(self, model_size: str = "default") -> None:
        """Load the Chatterbox multilingual model."""
@@ -103,132 +70,45 @@ class ChatterboxTTSBackend:

    def _load_model_sync(self):
        """Synchronous model loading."""
-        from ..utils.hf_progress import HFProgressTracker, create_hf_progress_callback
-
-        progress_manager = get_progress_manager()
-        task_manager = get_task_manager()
        model_name = "chatterbox-tts"
-
        is_cached = self._is_model_cached()

-        # Set up HF progress tracking (intercepts tqdm for file-level progress)
-        progress_callback = create_hf_progress_callback(model_name, progress_manager)
-        tracker = HFProgressTracker(progress_callback, filter_non_downloads=is_cached)
-        tracker_context = tracker.patch_download()
-        tracker_context.__enter__()
-
-        if not is_cached:
-            task_manager.start_download(model_name)
-            progress_manager.update_progress(
-                model_name=model_name,
-                current=0,
-                total=0,
-                filename="Connecting to HuggingFace...",
-                status="downloading",
-            )
-
-        try:
+        with model_load_progress(model_name, is_cached):
            device = self._get_device()
            self._device = device
-
            logger.info(f"Loading Chatterbox Multilingual TTS on {device}...")

            import torch
            from chatterbox.mtl_tts import ChatterboxMultilingualTTS

-            # Load into a local variable first, apply all patches, then
-            # assign to self.model.  This avoids leaving a half-initialised
-            # model on self.model if any patch step raises an exception.
-            #
-            # Monkey-patch torch.load for CPU loading. The model's .pt files
-            # were saved on CUDA; from_pretrained() doesn't pass map_location
-            # so loading on CPU fails without this.
-            try:
-                if device == "cpu":
-                    _orig_torch_load = torch.load
+            if device == "cpu":
+                _orig_torch_load = torch.load

-                    def _patched_load(*args, **kwargs):
-                        kwargs.setdefault("map_location", "cpu")
-                        return _orig_torch_load(*args, **kwargs)
+                def _patched_load(*args, **kwargs):
+                    kwargs.setdefault("map_location", "cpu")
+                    return _orig_torch_load(*args, **kwargs)

-                    with ChatterboxTTSBackend._load_lock:
-                        torch.load = _patched_load
-                        try:
-                            model = ChatterboxMultilingualTTS.from_pretrained(
-                                device=device,
-                            )
-                        finally:
-                            torch.load = _orig_torch_load
-                else:
-                    model = ChatterboxMultilingualTTS.from_pretrained(
-                        device=device,
-                    )
-            finally:
-                tracker_context.__exit__(None, None, None)
+                with ChatterboxTTSBackend._load_lock:
+                    torch.load = _patched_load
+                    try:
+                        model = ChatterboxMultilingualTTS.from_pretrained(device=device)
+                    finally:
+                        torch.load = _orig_torch_load
+            else:
+                model = ChatterboxMultilingualTTS.from_pretrained(device=device)

-            # Fix: transformers >= 4.36 defaults LlamaModel to sdpa attention
-            # which doesn't support output_attentions=True (needed by
-            # Chatterbox's AlignmentStreamAnalyzer). Force eager attention.
+            # Fix sdpa attention for output_attentions support
            t3_tfmr = model.t3.tfmr
-            if hasattr(t3_tfmr, "config") and hasattr(
-                t3_tfmr.config, "_attn_implementation"
-            ):
+            if hasattr(t3_tfmr, "config") and hasattr(t3_tfmr.config, "_attn_implementation"):
                t3_tfmr.config._attn_implementation = "eager"
                for layer in getattr(t3_tfmr, "layers", []):
                    if hasattr(layer, "self_attn"):
                        layer.self_attn._attn_implementation = "eager"

-            if not is_cached:
-                progress_manager.mark_complete(model_name)
-                task_manager.complete_download(model_name)
-
-            # Patch float64 → float32 dtype mismatches in upstream chatterbox.
-            # librosa.load returns float64 numpy; multiple upstream code paths
-            # convert it to a torch tensor via torch.from_numpy() without
-            # casting, then matmul it against float32 model weights.
-            import types
-
-            # Patch S3Tokenizer (used by s3gen.tokenizer)
-            _tokzr = model.s3gen.tokenizer
-            _orig_log_mel = _tokzr.log_mel_spectrogram.__func__
-
-            def _f32_log_mel(self_tokzr, audio, padding=0):
-                import torch as _torch
-                if _torch.is_tensor(audio):
-                    audio = audio.float()
-                return _orig_log_mel(self_tokzr, audio, padding)
-
-            _tokzr.log_mel_spectrogram = types.MethodType(_f32_log_mel, _tokzr)
-
-            # Patch VoiceEncoder
-            _ve = model.ve
-            _orig_ve_forward = _ve.forward.__func__
-
-            def _f32_ve_forward(self_ve, mels):
-                return _orig_ve_forward(self_ve, mels.float())
-
-            _ve.forward = types.MethodType(_f32_ve_forward, _ve)
-
-            # All patches applied successfully — publish the model
+            patch_chatterbox_f32(model)
            self.model = model

-            logger.info("Chatterbox Multilingual TTS loaded successfully")
-
-        except ImportError as e:
-            logger.error(
-                "chatterbox-tts package not found. "
-                "Install with: pip install chatterbox-tts"
-            )
-            if not is_cached:
-                progress_manager.mark_error(model_name, str(e))
-                task_manager.error_download(model_name, str(e))
-            raise
-        except Exception as e:
-            logger.error(f"Failed to load Chatterbox: {e}")
-            if not is_cached:
-                progress_manager.mark_error(model_name, str(e))
-                task_manager.error_download(model_name, str(e))
-            raise
+        logger.info("Chatterbox Multilingual TTS loaded successfully")

    def unload_model(self) -> None:
        """Unload model to free memory."""
@@ -267,17 +147,7 @@ class ChatterboxTTSBackend:
        audio_paths: List[str],
        reference_texts: List[str],
    ) -> Tuple[np.ndarray, str]:
-        """Combine multiple reference samples."""
-        combined_audio = []
-        for path in audio_paths:
-            audio, _sr = load_audio(path)
-            audio = normalize_audio(audio)
-            combined_audio.append(audio)
-
-        mixed = np.concatenate(combined_audio)
-        mixed = normalize_audio(mixed)
-        combined_text = " ".join(reference_texts)
-        return mixed, combined_text
+        return await _combine_voice_prompts(audio_paths, reference_texts)

    # Per-language generation defaults. Lower temp + higher cfg = clearer speech.
    _LANG_DEFAULTS: ClassVar[dict] = {
@@ -8,7 +8,6 @@ Forces CPU on macOS due to known MPS tensor issues.

 import asyncio
 import logging
-import platform
 import threading
 from pathlib import Path
 from typing import ClassVar, List, Optional, Tuple
@@ -16,9 +15,13 @@ from typing import ClassVar, List, Optional, Tuple
 import numpy as np

 from . import TTSBackend
-from ..utils.audio import normalize_audio, load_audio
-from ..utils.progress import get_progress_manager
-from ..utils.tasks import get_task_manager
+from .base import (
+    is_model_cached,
+    get_torch_device,
+    combine_voice_prompts as _combine_voice_prompts,
+    model_load_progress,
+    patch_chatterbox_f32,
+)

 logger = logging.getLogger(__name__)

@@ -45,17 +48,7 @@ class ChatterboxTurboTTSBackend:
        self._model_load_lock = asyncio.Lock()

    def _get_device(self) -> str:
-        """Get the best available device. Forces CPU on macOS (MPS issue)."""
-        if platform.system() == "Darwin":
-            return "cpu"
-        try:
-            import torch
-
-            if torch.cuda.is_available():
-                return "cuda"
-        except ImportError:
-            pass
-        return "cpu"
+        return get_torch_device(force_cpu_on_mac=True)

    def is_loaded(self) -> bool:
        return self.model is not None
@@ -64,33 +57,7 @@ class ChatterboxTurboTTSBackend:
        return CHATTERBOX_TURBO_HF_REPO

    def _is_model_cached(self, model_size: str = "default") -> bool:
-        """Check if the Chatterbox Turbo model is cached locally."""
-        try:
-            from huggingface_hub import constants as hf_constants
-
-            repo_cache = Path(hf_constants.HF_HUB_CACHE) / (
-                "models--" + CHATTERBOX_TURBO_HF_REPO.replace("/", "--")
-            )
-
-            if not repo_cache.exists():
-                return False
-
-            blobs_dir = repo_cache / "blobs"
-            if blobs_dir.exists() and any(blobs_dir.glob("*.incomplete")):
-                return False
-
-            # Check for turbo weight files
-            snapshots_dir = repo_cache / "snapshots"
-            if snapshots_dir.exists():
-                for fname in _TURBO_WEIGHT_FILES:
-                    if not any(snapshots_dir.rglob(fname)):
-                        return False
-                return True
-
-            return False
-        except Exception as e:
-            logger.warning(f"Error checking Chatterbox Turbo cache: {e}")
-            return False
+        return is_model_cached(CHATTERBOX_TURBO_HF_REPO, required_files=_TURBO_WEIGHT_FILES)

    async def load_model(self, model_size: str = "default") -> None:
        """Load the Chatterbox Turbo model."""
@@ -103,59 +70,24 @@ class ChatterboxTurboTTSBackend:

    def _load_model_sync(self):
        """Synchronous model loading."""
-        from ..utils.hf_progress import HFProgressTracker, create_hf_progress_callback
-
-        progress_manager = get_progress_manager()
-        task_manager = get_task_manager()
        model_name = "chatterbox-turbo"
-
        is_cached = self._is_model_cached()

-        # Set up HF progress tracking (intercepts tqdm for file-level progress)
-        progress_callback = create_hf_progress_callback(model_name, progress_manager)
-        tracker = HFProgressTracker(progress_callback, filter_non_downloads=is_cached)
-        tracker_context = tracker.patch_download()
-        tracker_context.__enter__()
-
-        if not is_cached:
-            task_manager.start_download(model_name)
-            progress_manager.update_progress(
-                model_name=model_name,
-                current=0,
-                total=0,
-                filename="Connecting to HuggingFace...",
-                status="downloading",
-            )
-
-        try:
+        with model_load_progress(model_name, is_cached):
            device = self._get_device()
            self._device = device
-
            logger.info(f"Loading Chatterbox Turbo TTS on {device}...")

            import torch
            from huggingface_hub import snapshot_download
            from chatterbox.tts_turbo import ChatterboxTurboTTS

-            # Download model files ourselves so we can pass token=None
-            # (upstream from_pretrained passes token=True which requires
-            # a stored HF token even though the repo is public).
-            try:
-                local_path = snapshot_download(
-                    repo_id=CHATTERBOX_TURBO_HF_REPO,
-                    token=None,
-                    allow_patterns=[
-                        "*.safetensors", "*.json", "*.txt", "*.pt", "*.model",
-                    ],
-                )
-            finally:
-                tracker_context.__exit__(None, None, None)
+            local_path = snapshot_download(
+                repo_id=CHATTERBOX_TURBO_HF_REPO,
+                token=None,
+                allow_patterns=["*.safetensors", "*.json", "*.txt", "*.pt", "*.model"],
+            )

-            # Monkey-patch torch.load for CPU loading. The model's .pt files
-            # were saved on CUDA; from_local() doesn't pass map_location
-            # so loading on CPU fails without this.
-            # Load into a local var, apply patches, then publish to
-            # self.model so a failed patch doesn't leave us half-initialised.
            if device == "cpu":
                _orig_torch_load = torch.load

@@ -166,73 +98,16 @@ class ChatterboxTurboTTSBackend:
                with ChatterboxTurboTTSBackend._load_lock:
                    torch.load = _patched_load
                    try:
-                        model = ChatterboxTurboTTS.from_local(
-                            local_path, device,
-                        )
+                        model = ChatterboxTurboTTS.from_local(local_path, device)
                    finally:
                        torch.load = _orig_torch_load
            else:
-                model = ChatterboxTurboTTS.from_local(
-                    local_path, device,
-                )
+                model = ChatterboxTurboTTS.from_local(local_path, device)

-            if not is_cached:
-                progress_manager.mark_complete(model_name)
-                task_manager.complete_download(model_name)
-
-            # Patch float64 → float32 dtype mismatches in upstream chatterbox.
-            # librosa.load returns float64 numpy; multiple upstream code paths
-            # convert it to a torch tensor via torch.from_numpy() without
-            # casting, then matmul it against float32 model weights.
-            # We patch the two known entry points:
-            #
-            # 1. S3Tokenizer.log_mel_spectrogram — the audio tensor from
-            #    librosa hits _mel_filters (float32) in a matmul.
-            # 2. VoiceEncoder.forward — float64 mel spectrograms hit the
-            #    float32 LSTM weights.
-            import types
-
-            # Patch S3Tokenizer (used by s3gen.tokenizer)
-            _tokzr = model.s3gen.tokenizer
-            _orig_log_mel = _tokzr.log_mel_spectrogram.__func__
-
-            def _f32_log_mel(self_tokzr, audio, padding=0):
-                import torch as _torch
-                if _torch.is_tensor(audio):
-                    audio = audio.float()
-                return _orig_log_mel(self_tokzr, audio, padding)
-
-            _tokzr.log_mel_spectrogram = types.MethodType(_f32_log_mel, _tokzr)
-
-            # Patch VoiceEncoder
-            _ve = model.ve
-            _orig_ve_forward = _ve.forward.__func__
-
-            def _f32_ve_forward(self_ve, mels):
-                return _orig_ve_forward(self_ve, mels.float())
-
-            _ve.forward = types.MethodType(_f32_ve_forward, _ve)
-
-            # Only publish after all patches succeed
+            patch_chatterbox_f32(model)
            self.model = model

-            logger.info("Chatterbox Turbo TTS loaded successfully")
-
-        except ImportError as e:
-            logger.error(
-                "chatterbox-tts package not found. "
-                "Install with: pip install chatterbox-tts"
-            )
-            if not is_cached:
-                progress_manager.mark_error(model_name, str(e))
-                task_manager.error_download(model_name, str(e))
-            raise
-        except Exception as e:
-            logger.error(f"Failed to load Chatterbox Turbo: {e}")
-            if not is_cached:
-                progress_manager.mark_error(model_name, str(e))
-                task_manager.error_download(model_name, str(e))
-            raise
+        logger.info("Chatterbox Turbo TTS loaded successfully")

    def unload_model(self) -> None:
        """Unload model to free memory."""
@@ -270,17 +145,7 @@ class ChatterboxTurboTTSBackend:
        audio_paths: List[str],
        reference_texts: List[str],
    ) -> Tuple[np.ndarray, str]:
-        """Combine multiple reference samples."""
-        combined_audio = []
-        for path in audio_paths:
-            audio, _sr = load_audio(path)
-            audio = normalize_audio(audio)
-            combined_audio.append(audio)
-
-        mixed = np.concatenate(combined_audio)
-        mixed = normalize_audio(mixed)
-        combined_text = " ".join(reference_texts)
-        return mixed, combined_text
+        return await _combine_voice_prompts(audio_paths, reference_texts)

    async def generate(
        self,
@@ -7,16 +7,13 @@ Wraps the LuxTTS (ZipVoice) model for zero-shot voice cloning.

 import asyncio
 import logging
-from pathlib import Path
-from typing import List, Optional, Tuple
+from typing import Optional, Tuple

 import numpy as np

 from . import TTSBackend
-from ..utils.audio import normalize_audio, load_audio
+from .base import is_model_cached, get_torch_device, combine_voice_prompts as _combine_voice_prompts, model_load_progress
 from ..utils.cache import get_cache_key, get_cached_voice_prompt, cache_voice_prompt
-from ..utils.progress import get_progress_manager
-from ..utils.tasks import get_task_manager

 logger = logging.getLogger(__name__)

@@ -33,14 +30,7 @@ class LuxTTSBackend:
        self._device = None

    def _get_device(self) -> str:
-        """Get the best available device."""
-        import torch
-
-        if torch.cuda.is_available():
-            return "cuda"
-        if hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
-            return "mps"
-        return "cpu"
+        return get_torch_device(allow_mps=True)

    def is_loaded(self) -> bool:
        return self.model is not None
@@ -55,35 +45,10 @@ class LuxTTSBackend:
        return LUXTTS_HF_REPO

    def _is_model_cached(self, model_size: str = "default") -> bool:
-        """Check if LuxTTS model weights are cached locally."""
-        try:
-            from huggingface_hub import constants as hf_constants
-
-            repo_cache = (
-                Path(hf_constants.HF_HUB_CACHE)
-                / ("models--" + LUXTTS_HF_REPO.replace("/", "--"))
-            )
-
-            if not repo_cache.exists():
-                return False
-
-            blobs_dir = repo_cache / "blobs"
-            if blobs_dir.exists() and any(blobs_dir.glob("*.incomplete")):
-                return False
-
-            snapshots_dir = repo_cache / "snapshots"
-            if snapshots_dir.exists():
-                has_weights = any(snapshots_dir.rglob("*.pt")) or any(
-                    snapshots_dir.rglob("*.safetensors")
-                ) or any(snapshots_dir.rglob("*.onnx")) or any(
-                    snapshots_dir.rglob("*.bin")
-                )
-                return has_weights
-
-            return False
-        except Exception as e:
-            logger.warning(f"Error checking LuxTTS cache: {e}")
-            return False
+        return is_model_cached(
+            LUXTTS_HF_REPO,
+            weight_extensions=(".pt", ".safetensors", ".onnx", ".bin"),
+        )

    async def load_model(self, model_size: str = "default") -> None:
        """Load the LuxTTS model."""
@@ -93,67 +58,25 @@ class LuxTTSBackend:
        await asyncio.to_thread(self._load_model_sync)

    def _load_model_sync(self):
-        """Synchronous model loading."""
-        from ..utils.hf_progress import HFProgressTracker, create_hf_progress_callback
-
-        progress_manager = get_progress_manager()
-        task_manager = get_task_manager()
        model_name = "luxtts"
-
        is_cached = self._is_model_cached()

-        # Set up HF progress tracking (intercepts tqdm for file-level progress)
-        progress_callback = create_hf_progress_callback(model_name, progress_manager)
-        tracker = HFProgressTracker(progress_callback, filter_non_downloads=is_cached)
-        tracker_context = tracker.patch_download()
-        tracker_context.__enter__()
-
-        if not is_cached:
-            task_manager.start_download(model_name)
-            progress_manager.update_progress(
-                model_name=model_name,
-                current=0,
-                total=0,
-                filename="Connecting to HuggingFace...",
-                status="downloading",
-            )
-
-        try:
+        with model_load_progress(model_name, is_cached):
            from zipvoice.luxvoice import LuxTTS

            device = self.device
            logger.info(f"Loading LuxTTS on {device}...")

-            # LuxTTS constructor downloads model and loads everything
-            try:
-                if device == "cpu":
-                    import os
-                    threads = os.cpu_count() or 4
-                    self.model = LuxTTS(
-                        model_path=LUXTTS_HF_REPO,
-                        device="cpu",
-                        threads=min(threads, 8),
-                    )
-                else:
-                    self.model = LuxTTS(
-                        model_path=LUXTTS_HF_REPO,
-                        device=device,
-                    )
-            finally:
-                tracker_context.__exit__(None, None, None)
+            if device == "cpu":
+                import os
+                threads = os.cpu_count() or 4
+                self.model = LuxTTS(
+                    model_path=LUXTTS_HF_REPO, device="cpu", threads=min(threads, 8),
+                )
+            else:
+                self.model = LuxTTS(model_path=LUXTTS_HF_REPO, device=device)

-            if not is_cached:
-                progress_manager.mark_complete(model_name)
-                task_manager.complete_download(model_name)
-
-            logger.info("LuxTTS loaded successfully")
-
-        except Exception as e:
-            logger.error(f"Failed to load LuxTTS: {e}")
-            if not is_cached:
-                progress_manager.mark_error(model_name, str(e))
-                task_manager.error_download(model_name, str(e))
-            raise
+        logger.info("LuxTTS loaded successfully")

    def unload_model(self) -> None:
        """Unload model to free memory."""
@@ -204,28 +127,8 @@ class LuxTTSBackend:

        return encoded, False

-    async def combine_voice_prompts(
-        self,
-        audio_paths: List[str],
-        reference_texts: List[str],
-    ) -> Tuple[np.ndarray, str]:
-        """
-        Combine multiple reference samples.
-
-        LuxTTS doesn't have native multi-prompt support, so we concatenate
-        the audio and let encode_prompt handle the combined clip.
-        """
-        combined_audio = []
-        for path in audio_paths:
-            audio, _sr = load_audio(path, sample_rate=24000)
-            audio = normalize_audio(audio)
-            combined_audio.append(audio)
-
-        mixed = np.concatenate(combined_audio)
-        mixed = normalize_audio(mixed)
-        combined_text = " ".join(reference_texts)
-
-        return mixed, combined_text
+    async def combine_voice_prompts(self, audio_paths, reference_texts):
+        return await _combine_voice_prompts(audio_paths, reference_texts, sample_rate=24000)

    async def generate(
        self,
@@ -4,49 +4,44 @@ MLX backend implementation for TTS and STT using mlx-audio.

 from typing import Optional, List, Tuple
 import asyncio
+import logging
 import numpy as np
 import os
 from pathlib import Path

+logger = logging.getLogger(__name__)
+
 # PATCH: Import and apply offline patch BEFORE any huggingface_hub usage
 # This prevents mlx_audio from making network requests when models are cached
 from ..utils.hf_offline_patch import patch_huggingface_hub_offline, ensure_original_qwen_config_cached
+
 patch_huggingface_hub_offline()
 ensure_original_qwen_config_cached()

-from . import TTSBackend, STTBackend
+from . import TTSBackend, STTBackend, LANGUAGE_CODE_TO_NAME, WHISPER_HF_REPOS
+from .base import is_model_cached, combine_voice_prompts as _combine_voice_prompts, model_load_progress
 from ..utils.cache import get_cache_key, get_cached_voice_prompt, cache_voice_prompt
-from ..utils.audio import normalize_audio, load_audio
-from ..utils.progress import get_progress_manager
-from ..utils.hf_progress import HFProgressTracker, create_hf_progress_callback
-from ..utils.tasks import get_task_manager
-
-LANGUAGE_CODE_TO_NAME = {
-    "zh": "chinese", "en": "english", "ja": "japanese", "ko": "korean",
-    "de": "german", "fr": "french", "ru": "russian", "pt": "portuguese",
-    "es": "spanish", "it": "italian",
-}


 class MLXTTSBackend:
    """MLX-based TTS backend using mlx-audio."""
-    
+
    def __init__(self, model_size: str = "1.7B"):
        self.model = None
        self.model_size = model_size
        self._current_model_size = None
-    
+
    def is_loaded(self) -> bool:
        """Check if model is loaded."""
        return self.model is not None
-    
+
    def _get_model_path(self, model_size: str) -> str:
        """
        Get the MLX model path.
-        
+
        Args:
            model_size: Model size (1.7B or 0.6B)
-            
+
        Returns:
            HuggingFace Hub model ID for MLX
        """
@@ -56,187 +51,90 @@ class MLXTTSBackend:
            # 0.6B not yet converted to MLX format
            "0.6B": "mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16",  # Fallback to 1.7B
        }
-        
+
        if model_size not in mlx_model_map:
            raise ValueError(f"Unknown model size: {model_size}")
-        
+
        hf_model_id = mlx_model_map[model_size]
-        print(f"Will download MLX model from HuggingFace Hub: {hf_model_id}")
-        
+        logger.info("Will download MLX model from HuggingFace Hub: %s", hf_model_id)
+
        return hf_model_id
-    
+
    def _is_model_cached(self, model_size: str) -> bool:
-        """
-        Check if the model is already cached locally AND fully downloaded.
-        
-        Args:
-            model_size: Model size to check
-            
-        Returns:
-            True if model is fully cached, False if missing or incomplete
-        """
-        try:
-            from huggingface_hub import constants as hf_constants
-            model_path = self._get_model_path(model_size)
-            repo_cache = Path(hf_constants.HF_HUB_CACHE) / ("models--" + model_path.replace("/", "--"))
-            
-            if not repo_cache.exists():
-                return False
-            
-            # Check for .incomplete files - if any exist, download is still in progress
-            blobs_dir = repo_cache / "blobs"
-            if blobs_dir.exists() and any(blobs_dir.glob("*.incomplete")):
-                print(f"[_is_model_cached] Found .incomplete files for {model_size}, treating as not cached")
-                return False
-            
-            # Check that actual model weight files exist in snapshots
-            snapshots_dir = repo_cache / "snapshots"
-            if snapshots_dir.exists():
-                has_weights = (
-                    any(snapshots_dir.rglob("*.safetensors")) or
-                    any(snapshots_dir.rglob("*.bin")) or
-                    any(snapshots_dir.rglob("*.npz"))
-                )
-                if not has_weights:
-                    print(f"[_is_model_cached] No model weights found for {model_size}, treating as not cached")
-                    return False
-            
-            return True
-        except Exception as e:
-            print(f"[_is_model_cached] Error checking cache for {model_size}: {e}")
-            return False
-    
+        return is_model_cached(
+            self._get_model_path(model_size),
+            weight_extensions=(".safetensors", ".bin", ".npz"),
+        )
+
    async def load_model_async(self, model_size: Optional[str] = None):
        """
        Lazy load the MLX TTS model.
-        
+
        Args:
            model_size: Model size to load (1.7B or 0.6B)
        """
        if model_size is None:
            model_size = self.model_size
-            
+
        # If already loaded with correct size, return
        if self.model is not None and self._current_model_size == model_size:
            return
-        
+
        # Unload existing model if different size requested
        if self.model is not None and self._current_model_size != model_size:
            self.unload_model()
-        
+
        # Run blocking load in thread pool
        await asyncio.to_thread(self._load_model_sync, model_size)
-    
+
    # Alias for compatibility
    load_model = load_model_async
-    
+
    def _load_model_sync(self, model_size: str):
        """Synchronous model loading."""
+        model_path = self._get_model_path(model_size)
+        model_name = f"qwen-tts-{model_size}"
+        is_cached = self._is_model_cached(model_size)
+
+        # Force offline mode when cached to avoid network requests
+        original_hf_hub_offline = os.environ.get("HF_HUB_OFFLINE")
+        if is_cached:
+            os.environ["HF_HUB_OFFLINE"] = "1"
+            logger.info("[PATCH] Model %s is cached, forcing HF_HUB_OFFLINE=1 to avoid network requests", model_size)
+
        try:
-            # Get model path BEFORE importing mlx_audio
-            model_path = self._get_model_path(model_size)
-            
-            # Set up progress tracking
-            progress_manager = get_progress_manager()
-            task_manager = get_task_manager()
-            model_name = f"qwen-tts-{model_size}"
-            
-            # Check if model is already cached
-            is_cached = self._is_model_cached(model_size)
-            
-            # Set up progress callback
-            # If cached: filter out non-download progress
-            # If not cached: report all progress (we're actually downloading)
-            progress_callback = create_hf_progress_callback(model_name, progress_manager)
-            tracker = HFProgressTracker(progress_callback, filter_non_downloads=is_cached)
-            
-            print(f"Loading MLX TTS model {model_size}...")
-            
-            # Only track download progress if model is NOT cached
-            if not is_cached:
-                # Start tracking download task
-                task_manager.start_download(model_name)
-                
-                # Initialize progress state so SSE endpoint has initial data to send
-                # This provides immediate feedback while HuggingFace fetches metadata
-                progress_manager.update_progress(
-                    model_name=model_name,
-                    current=0,
-                    total=0,  # Will be updated once actual total is known
-                    filename="Connecting to HuggingFace...",
-                    status="downloading",
-                )
-            
-            # IMPORTANT: Patch tqdm BEFORE importing mlx_audio
-            # Otherwise mlx_audio caches reference to original tqdm
-            tracker_context = tracker.patch_download()
-            tracker_context.__enter__()
-            
-            # PATCH: Force offline mode when model is already cached
-            # This prevents crashes when HuggingFace is unreachable
-            original_hf_hub_offline = os.environ.get("HF_HUB_OFFLINE")
-            if is_cached:
-                os.environ["HF_HUB_OFFLINE"] = "1"
-                print(f"[PATCH] Model {model_size} is cached, forcing HF_HUB_OFFLINE=1 to avoid network requests")
-            
-            # Import mlx_audio AFTER patching tqdm
-            from mlx_audio.tts import load
-            
-            # Load MLX model (downloads automatically)
-            try:
-                self.model = load(model_path)
-            except Exception as load_error:
-                # If offline mode failed, try with network enabled as fallback
-                if is_cached and "offline" in str(load_error).lower():
-                    print(f"[PATCH] Offline load failed, trying with network: {load_error}")
-                    os.environ.pop("HF_HUB_OFFLINE", None)
+            with model_load_progress(model_name, is_cached):
+                from mlx_audio.tts import load
+
+                logger.info("Loading MLX TTS model %s...", model_size)
+
+                try:
                    self.model = load(model_path)
-                else:
-                    raise
-            finally:
-                # Exit the patch context
-                tracker_context.__exit__(None, None, None)
-                # Restore original HF_HUB_OFFLINE setting
-                if original_hf_hub_offline is not None:
-                    os.environ["HF_HUB_OFFLINE"] = original_hf_hub_offline
-                else:
-                    os.environ.pop("HF_HUB_OFFLINE", None)
-            
-            # Only mark download as complete if we were tracking it
-            if not is_cached:
-                progress_manager.mark_complete(model_name)
-                task_manager.complete_download(model_name)
-            
-            self._current_model_size = model_size
-            self.model_size = model_size
-            
-            print(f"MLX TTS model {model_size} loaded successfully")
-            
-        except ImportError as e:
-            print(f"Error: mlx_audio package not found. Install with: pip install mlx-audio")
-            progress_manager = get_progress_manager()
-            task_manager = get_task_manager()
-            model_name = f"qwen-tts-{model_size}"
-            progress_manager.mark_error(model_name, str(e))
-            task_manager.error_download(model_name, str(e))
-            raise
-        except Exception as e:
-            print(f"Error loading MLX TTS model: {e}")
-            progress_manager = get_progress_manager()
-            task_manager = get_task_manager()
-            model_name = f"qwen-tts-{model_size}"
-            progress_manager.mark_error(model_name, str(e))
-            task_manager.error_download(model_name, str(e))
-            raise
-    
+                except Exception as load_error:
+                    if is_cached and "offline" in str(load_error).lower():
+                        logger.warning("[PATCH] Offline load failed, trying with network: %s", load_error)
+                        os.environ.pop("HF_HUB_OFFLINE", None)
+                        self.model = load(model_path)
+                    else:
+                        raise
+        finally:
+            if original_hf_hub_offline is not None:
+                os.environ["HF_HUB_OFFLINE"] = original_hf_hub_offline
+            else:
+                os.environ.pop("HF_HUB_OFFLINE", None)
+
+        self._current_model_size = model_size
+        self.model_size = model_size
+        logger.info("MLX TTS model %s loaded successfully", model_size)
+
    def unload_model(self):
        """Unload the model to free memory."""
        if self.model is not None:
            del self.model
            self.model = None
            self._current_model_size = None
-            print("MLX TTS model unloaded")
-    
+            logger.info("MLX TTS model unloaded")
+
    async def create_voice_prompt(
        self,
        audio_path: str,
@@ -245,20 +143,20 @@ class MLXTTSBackend:
    ) -> Tuple[dict, bool]:
        """
        Create voice prompt from reference audio.
-        
+
        MLX backend stores voice prompt as a dict with audio path and text.
        The actual voice prompt processing happens during generation.
-        
+
        Args:
            audio_path: Path to reference audio file
            reference_text: Transcript of reference audio
            use_cache: Whether to use cached prompt if available
-            
+
        Returns:
            Tuple of (voice_prompt_dict, was_cached)
        """
        await self.load_model_async(None)
-        
+
        # Check cache if enabled
        if use_cache:
            cache_key = get_cache_key(audio_path, reference_text)
@@ -272,53 +170,25 @@ class MLXTTSBackend:
                        return cached_prompt, True
                    else:
                        # Cached file no longer exists, invalidate cache
-                        print(f"Cached audio file not found: {cached_audio_path}, regenerating prompt")
-        
+                        logger.warning("Cached audio file not found: %s, regenerating prompt", cached_audio_path)
+
        # MLX voice prompt format - store audio path and text
        # The model will process this during generation
        voice_prompt_items = {
            "ref_audio": str(audio_path),
            "ref_text": reference_text,
        }
-        
+
        # Cache if enabled
        if use_cache:
            cache_key = get_cache_key(audio_path, reference_text)
            cache_voice_prompt(cache_key, voice_prompt_items)
-        
+
        return voice_prompt_items, False
-    
-    async def combine_voice_prompts(
-        self,
-        audio_paths: List[str],
-        reference_texts: List[str],
-    ) -> Tuple[np.ndarray, str]:
-        """
-        Combine multiple reference samples for better quality.
-        
-        Args:
-            audio_paths: List of audio file paths
-            reference_texts: List of reference texts
-            
-        Returns:
-            Tuple of (combined_audio, combined_text)
-        """
-        combined_audio = []
-        
-        for audio_path in audio_paths:
-            audio, sr = load_audio(audio_path)
-            audio = normalize_audio(audio)
-            combined_audio.append(audio)
-        
-        # Concatenate audio
-        mixed = np.concatenate(combined_audio)
-        mixed = normalize_audio(mixed)
-        
-        # Combine texts
-        combined_text = " ".join(reference_texts)
-        
-        return mixed, combined_text
-    
+
+    async def combine_voice_prompts(self, audio_paths, reference_texts):
+        return await _combine_voice_prompts(audio_paths, reference_texts)
+
    async def generate(
        self,
        text: str,
@@ -342,7 +212,7 @@ class MLXTTSBackend:
        """
        await self.load_model_async(None)

-        print(f"Generating audio for text: {text}")
+        logger.info("Generating audio for text: %s", text)

        def _generate_sync():
            """Run synchronous generation in thread pool."""
@@ -354,20 +224,21 @@ class MLXTTSBackend:
            # Set seed if provided (MLX uses numpy random)
            if seed is not None:
                import mlx.core as mx
+
                np.random.seed(seed)
                mx.random.seed(seed)
-            
+
            # Extract voice prompt info
            ref_audio = voice_prompt.get("ref_audio") or voice_prompt.get("ref_audio_path")
            ref_text = voice_prompt.get("ref_text", "")
-            
+
            # Validate that the audio file exists
            if ref_audio and not Path(ref_audio).exists():
-                print(f"Warning: Audio file not found: {ref_audio}")
-                print("This may be due to a cached voice prompt referencing a deleted temp file.")
-                print("Regenerating without voice prompt.")
+                logger.warning("Audio file not found: %s", ref_audio)
+                logger.warning("This may be due to a cached voice prompt referencing a deleted temp file.")
+                logger.warning("Regenerating without voice prompt.")
                ref_audio = None
-            
+
            # Check if model supports voice cloning via generate method
            # MLX API may support ref_audio parameter directly
            try:
@@ -375,6 +246,7 @@ class MLXTTSBackend:
                if ref_audio:
                    # Check if generate accepts ref_audio parameter
                    import inspect
+
                    sig = inspect.signature(self.model.generate)
                    if "ref_audio" in sig.parameters:
                        # Generate with voice cloning
@@ -393,18 +265,18 @@ class MLXTTSBackend:
                        sample_rate = result.sample_rate
            except Exception as e:
                # If voice cloning fails, try without it
-                print(f"Warning: Voice cloning failed, generating without voice prompt: {e}")
+                logger.warning("Voice cloning failed, generating without voice prompt: %s", e)
                for result in self.model.generate(text, lang_code=lang):
                    audio_chunks.append(np.array(result.audio))
                    sample_rate = result.sample_rate
-            
+
            # Concatenate all chunks
            if audio_chunks:
                audio = np.concatenate([np.asarray(chunk, dtype=np.float32) for chunk in audio_chunks])
            else:
                # Fallback: empty audio
                audio = np.array([], dtype=np.float32)
-            
+
            return audio, sample_rate

        # Run blocking inference in thread pool
@@ -413,167 +285,62 @@ class MLXTTSBackend:
        return audio, sample_rate


-WHISPER_HF_REPOS = {
-    "base": "openai/whisper-base",
-    "small": "openai/whisper-small",
-    "medium": "openai/whisper-medium",
-    "large": "openai/whisper-large-v3",
-}
-
-
 class MLXSTTBackend:
    """MLX-based STT backend using mlx-audio Whisper."""

    def __init__(self, model_size: str = "base"):
        self.model = None
        self.model_size = model_size
-    
+
    def is_loaded(self) -> bool:
        """Check if model is loaded."""
        return self.model is not None
-    
+
    def _is_model_cached(self, model_size: str) -> bool:
-        """
-        Check if the Whisper model is already cached locally AND fully downloaded.
-        
-        Args:
-            model_size: Model size to check
-            
-        Returns:
-            True if model is fully cached, False if missing or incomplete
-        """
-        try:
-            from huggingface_hub import constants as hf_constants
-            hf_repo = WHISPER_HF_REPOS.get(model_size, f"openai/whisper-{model_size}")
-            repo_cache = Path(hf_constants.HF_HUB_CACHE) / ("models--" + hf_repo.replace("/", "--"))
-            
-            if not repo_cache.exists():
-                return False
-            
-            # Check for .incomplete files - if any exist, download is still in progress
-            blobs_dir = repo_cache / "blobs"
-            if blobs_dir.exists() and any(blobs_dir.glob("*.incomplete")):
-                print(f"[_is_model_cached] Found .incomplete files for whisper-{model_size}, treating as not cached")
-                return False
-            
-            # Check that actual model weight files exist in snapshots
-            snapshots_dir = repo_cache / "snapshots"
-            if snapshots_dir.exists():
-                has_weights = (
-                    any(snapshots_dir.rglob("*.safetensors")) or
-                    any(snapshots_dir.rglob("*.bin")) or
-                    any(snapshots_dir.rglob("*.npz"))
-                )
-                if not has_weights:
-                    print(f"[_is_model_cached] No model weights found for whisper-{model_size}, treating as not cached")
-                    return False
-            
-            return True
-        except Exception as e:
-            print(f"[_is_model_cached] Error checking cache for whisper-{model_size}: {e}")
-            return False
-    
+        hf_repo = WHISPER_HF_REPOS.get(model_size, f"openai/whisper-{model_size}")
+        return is_model_cached(hf_repo, weight_extensions=(".safetensors", ".bin", ".npz"))
+
    async def load_model_async(self, model_size: Optional[str] = None):
        """
        Lazy load the MLX Whisper model.
-        
+
        Args:
            model_size: Model size (tiny, base, small, medium, large)
        """
        if model_size is None:
            model_size = self.model_size
-        
+
        if self.model is not None and self.model_size == model_size:
            return
-        
+
        # Run blocking load in thread pool
        await asyncio.to_thread(self._load_model_sync, model_size)
-    
+
    # Alias for compatibility
    load_model = load_model_async
-    
+
    def _load_model_sync(self, model_size: str):
        """Synchronous model loading."""
-        try:
-            progress_manager = get_progress_manager()
-            task_manager = get_task_manager()
-            progress_model_name = f"whisper-{model_size}"
+        progress_model_name = f"whisper-{model_size}"
+        is_cached = self._is_model_cached(model_size)

-            # Check if model is already cached
-            is_cached = self._is_model_cached(model_size)
-
-            # Set up progress callback and tracker
-            # If cached: filter out non-download progress
-            # If not cached: report all progress (we're actually downloading)
-            progress_callback = create_hf_progress_callback(progress_model_name, progress_manager)
-            tracker = HFProgressTracker(progress_callback, filter_non_downloads=is_cached)
-
-            # Patch tqdm BEFORE importing mlx_audio
-            tracker_context = tracker.patch_download()
-            tracker_context.__enter__()
-
-            # Import mlx_audio
+        with model_load_progress(progress_model_name, is_cached):
            from mlx_audio.stt import load

-            # MLX Whisper uses the standard OpenAI models
            model_name = WHISPER_HF_REPOS.get(model_size, f"openai/whisper-{model_size}")
+            logger.info("Loading MLX Whisper model %s...", model_size)
+            self.model = load(model_name)

-            print(f"Loading MLX Whisper model {model_size}...")
+        self.model_size = model_size
+        logger.info("MLX Whisper model %s loaded successfully", model_size)

-            # Only track download progress if model is NOT cached
-            if not is_cached:
-                # Start tracking download task
-                task_manager.start_download(progress_model_name)
-                
-                # Initialize progress state so SSE endpoint has initial data to send
-                progress_manager.update_progress(
-                    model_name=progress_model_name,
-                    current=0,
-                    total=0,
-                    filename="Connecting to HuggingFace...",
-                    status="downloading",
-                )
-
-            # Load the model (tqdm is patched, but filters out non-download progress)
-            try:
-                self.model = load(model_name)
-            finally:
-                # Exit the patch context
-                tracker_context.__exit__(None, None, None)
-            
-            # Only mark download as complete if we were tracking it
-            if not is_cached:
-                progress_manager.mark_complete(progress_model_name)
-                task_manager.complete_download(progress_model_name)
-            
-            self.model_size = model_size
-            
-            print(f"MLX Whisper model {model_size} loaded successfully")
-            
-        except ImportError as e:
-            print(f"Error: mlx_audio package not found. Install with: pip install mlx-audio")
-            progress_manager = get_progress_manager()
-            task_manager = get_task_manager()
-            progress_model_name = f"whisper-{model_size}"
-            progress_manager.mark_error(progress_model_name, str(e))
-            task_manager.error_download(progress_model_name, str(e))
-            raise
-        except Exception as e:
-            print(f"Error loading MLX Whisper model: {e}")
-            progress_manager = get_progress_manager()
-            task_manager = get_task_manager()
-            progress_model_name = f"whisper-{model_size}"
-            progress_manager.mark_error(progress_model_name, str(e))
-            task_manager.error_download(progress_model_name, str(e))
-            raise
-    
    def unload_model(self):
        """Unload the model to free memory."""
        if self.model is not None:
            del self.model
            self.model = None
-            print("MLX Whisper model unloaded")
-    
+            logger.info("MLX Whisper model unloaded")
+
    async def transcribe(
        self,
        audio_path: str,
@@ -4,67 +4,47 @@ PyTorch backend implementation for TTS and STT.

 from typing import Optional, List, Tuple
 import asyncio
+import logging
 import torch
 import numpy as np
-from pathlib import Path

-from . import TTSBackend, STTBackend
+logger = logging.getLogger(__name__)
+
+from . import TTSBackend, STTBackend, LANGUAGE_CODE_TO_NAME, WHISPER_HF_REPOS
+from .base import (
+    is_model_cached,
+    get_torch_device,
+    combine_voice_prompts as _combine_voice_prompts,
+    model_load_progress,
+)
 from ..utils.cache import get_cache_key, get_cached_voice_prompt, cache_voice_prompt
-from ..utils.audio import normalize_audio, load_audio
-from ..utils.progress import get_progress_manager
-from ..utils.hf_progress import HFProgressTracker, create_hf_progress_callback
-from ..utils.tasks import get_task_manager
-
-LANGUAGE_CODE_TO_NAME = {
-    "zh": "chinese", "en": "english", "ja": "japanese", "ko": "korean",
-    "de": "german", "fr": "french", "ru": "russian", "pt": "portuguese",
-    "es": "spanish", "it": "italian",
-}
+from ..utils.audio import load_audio


 class PyTorchTTSBackend:
    """PyTorch-based TTS backend using Qwen3-TTS."""
-    
+
    def __init__(self, model_size: str = "1.7B"):
        self.model = None
        self.model_size = model_size
        self.device = self._get_device()
        self._current_model_size = None
-    
+
    def _get_device(self) -> str:
        """Get the best available device."""
-        if torch.cuda.is_available():
-            return "cuda"
-        # Intel Arc / Intel Xe GPU via intel-extension-for-pytorch (IPEX)
-        try:
-            import intel_extension_for_pytorch  # noqa: F401
-            if hasattr(torch, 'xpu') and torch.xpu.is_available():
-                return "xpu"
-        except ImportError:
-            pass
-        # Any GPU on Windows via DirectML (torch-directml)
-        try:
-            import torch_directml
-            if torch_directml.device_count() > 0:
-                return torch_directml.device(0)
-        except ImportError:
-            pass
-        # MPS (Apple Silicon) — kept for completeness but MLX backend is preferred
-        if hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
-            return "cpu"  # MPS disabled for stability; MLX backend handles Apple Silicon
-        return "cpu"
-    
+        return get_torch_device(allow_xpu=True, allow_directml=True)
+
    def is_loaded(self) -> bool:
        """Check if model is loaded."""
        return self.model is not None
-    
+
    def _get_model_path(self, model_size: str) -> str:
        """
        Get the HuggingFace Hub model ID.
-        
+
        Args:
            model_size: Model size (1.7B or 0.6B)
-            
+
        Returns:
            HuggingFace Hub model ID
        """
@@ -72,179 +52,79 @@ class PyTorchTTSBackend:
            "1.7B": "Qwen/Qwen3-TTS-12Hz-1.7B-Base",
            "0.6B": "Qwen/Qwen3-TTS-12Hz-0.6B-Base",
        }
-        
+
        if model_size not in hf_model_map:
            raise ValueError(f"Unknown model size: {model_size}")
-        
+
        return hf_model_map[model_size]
-    
+
    def _is_model_cached(self, model_size: str) -> bool:
-        """
-        Check if the model is already cached locally AND fully downloaded.
-        
-        Args:
-            model_size: Model size to check
-            
-        Returns:
-            True if model is fully cached, False if missing or incomplete
-        """
-        try:
-            from huggingface_hub import constants as hf_constants
-            model_path = self._get_model_path(model_size)
-            repo_cache = Path(hf_constants.HF_HUB_CACHE) / ("models--" + model_path.replace("/", "--"))
-            
-            if not repo_cache.exists():
-                return False
-            
-            # Check for .incomplete files - if any exist, download is still in progress
-            blobs_dir = repo_cache / "blobs"
-            if blobs_dir.exists() and any(blobs_dir.glob("*.incomplete")):
-                print(f"[_is_model_cached] Found .incomplete files for {model_size}, treating as not cached")
-                return False
-            
-            # Check that actual model weight files exist in snapshots
-            snapshots_dir = repo_cache / "snapshots"
-            if snapshots_dir.exists():
-                has_weights = (
-                    any(snapshots_dir.rglob("*.safetensors")) or
-                    any(snapshots_dir.rglob("*.bin"))
-                )
-                if not has_weights:
-                    print(f"[_is_model_cached] No model weights found for {model_size}, treating as not cached")
-                    return False
-            
-            return True
-        except Exception as e:
-            print(f"[_is_model_cached] Error checking cache for {model_size}: {e}")
-            return False
-    
+        return is_model_cached(self._get_model_path(model_size))
+
    async def load_model_async(self, model_size: Optional[str] = None):
        """
        Lazy load the TTS model with automatic downloading from HuggingFace Hub.
-        
+
        Args:
            model_size: Model size to load (1.7B or 0.6B)
        """
        if model_size is None:
            model_size = self.model_size
-            
+
        # If already loaded with correct size, return
        if self.model is not None and self._current_model_size == model_size:
            return
-        
+
        # Unload existing model if different size requested
        if self.model is not None and self._current_model_size != model_size:
            self.unload_model()
-        
+
        # Run blocking load in thread pool
        await asyncio.to_thread(self._load_model_sync, model_size)
-    
+
    # Alias for compatibility
    load_model = load_model_async
-    
+
    def _load_model_sync(self, model_size: str):
        """Synchronous model loading."""
-        try:
-            progress_manager = get_progress_manager()
-            task_manager = get_task_manager()
-            model_name = f"qwen-tts-{model_size}"
+        model_name = f"qwen-tts-{model_size}"
+        is_cached = self._is_model_cached(model_size)

-            # Check if model is already cached
-            is_cached = self._is_model_cached(model_size)
-
-            # Set up progress callback and tracker
-            # If cached: filter out non-download progress (like "Segment 1/1" during generation)
-            # If not cached: report all progress (we're actually downloading)
-            progress_callback = create_hf_progress_callback(model_name, progress_manager)
-            tracker = HFProgressTracker(progress_callback, filter_non_downloads=is_cached)
-
-            # Patch tqdm BEFORE importing qwen_tts
-            tracker_context = tracker.patch_download()
-            tracker_context.__enter__()
-
-            # Import qwen_tts
+        with model_load_progress(model_name, is_cached):
            from qwen_tts import Qwen3TTSModel

-            # Get model path (local or HuggingFace Hub ID)
            model_path = self._get_model_path(model_size)
+            logger.info("Loading TTS model %s on %s...", model_size, self.device)

-            print(f"Loading TTS model {model_size} on {self.device}...")
-
-            # Only track download progress if model is NOT cached
-            if not is_cached:
-                # Start tracking download task
-                task_manager.start_download(model_name)
-
-                # Initialize progress state so SSE endpoint has initial data to send
-                progress_manager.update_progress(
-                    model_name=model_name,
-                    current=0,
-                    total=0,  # Will be updated once actual total is known
-                    filename="Connecting to HuggingFace...",
-                    status="downloading",
+            if self.device == "cpu":
+                self.model = Qwen3TTSModel.from_pretrained(
+                    model_path,
+                    torch_dtype=torch.float32,
+                    low_cpu_mem_usage=False,
+                )
+            else:
+                self.model = Qwen3TTSModel.from_pretrained(
+                    model_path,
+                    device_map=self.device,
+                    torch_dtype=torch.bfloat16,
                )

-            # Load the model (tqdm is patched, but filters out non-download progress)
-            try:
-                # Don't pass device_map on CPU: accelerate's meta-tensor mechanism
-                # causes "Cannot copy out of meta tensor" when moving to CPU.
-                # Instead load directly then call .to(device) if needed.
-                if self.device == "cpu":
-                    self.model = Qwen3TTSModel.from_pretrained(
-                        model_path,
-                        torch_dtype=torch.float32,
-                        low_cpu_mem_usage=False,
-                    )
-                else:
-                    self.model = Qwen3TTSModel.from_pretrained(
-                        model_path,
-                        device_map=self.device,
-                        torch_dtype=torch.bfloat16,
-                    )
-            finally:
-                # Exit the patch context
-                tracker_context.__exit__(None, None, None)
-            
-            # Only mark download as complete if we were tracking it
-            if not is_cached:
-                progress_manager.mark_complete(model_name)
-                task_manager.complete_download(model_name)
-            
-            self._current_model_size = model_size
-            self.model_size = model_size
-            
-            print(f"TTS model {model_size} loaded successfully")
-            
-        except ImportError as e:
-            print(f"Error: qwen_tts package not found. Install with: pip install git+https://github.com/QwenLM/Qwen3-TTS.git")
-            progress_manager = get_progress_manager()
-            task_manager = get_task_manager()
-            model_name = f"qwen-tts-{model_size}"
-            progress_manager.mark_error(model_name, str(e))
-            task_manager.error_download(model_name, str(e))
-            raise
-        except Exception as e:
-            print(f"Error loading TTS model: {e}")
-            print(f"Tip: The model will be automatically downloaded from HuggingFace Hub on first use.")
-            progress_manager = get_progress_manager()
-            task_manager = get_task_manager()
-            model_name = f"qwen-tts-{model_size}"
-            progress_manager.mark_error(model_name, str(e))
-            task_manager.error_download(model_name, str(e))
-            raise
-    
+        self._current_model_size = model_size
+        self.model_size = model_size
+        logger.info("TTS model %s loaded successfully", model_size)
+
    def unload_model(self):
        """Unload the model to free memory."""
        if self.model is not None:
            del self.model
            self.model = None
            self._current_model_size = None
-            
+
            if torch.cuda.is_available():
                torch.cuda.empty_cache()
-            
-            print("TTS model unloaded")
-    
+
+            logger.info("TTS model unloaded")
+
    async def create_voice_prompt(
        self,
        audio_path: str,
@@ -253,17 +133,17 @@ class PyTorchTTSBackend:
    ) -> Tuple[dict, bool]:
        """
        Create voice prompt from reference audio.
-        
+
        Args:
            audio_path: Path to reference audio file
            reference_text: Transcript of reference audio
            use_cache: Whether to use cached prompt if available
-            
+
        Returns:
            Tuple of (voice_prompt_dict, was_cached)
        """
        await self.load_model_async(None)
-        
+
        # Check cache if enabled
        if use_cache:
            cache_key = get_cache_key(audio_path, reference_text)
@@ -279,7 +159,7 @@ class PyTorchTTSBackend:
                    # Legacy cache format - convert to dict
                    # This shouldn't happen in practice, but handle it
                    return {"prompt": cached_prompt}, True
-        
+
        def _create_prompt_sync():
            """Run synchronous voice prompt creation in thread pool."""
            return self.model.create_voice_clone_prompt(
@@ -287,48 +167,24 @@ class PyTorchTTSBackend:
                ref_text=reference_text,
                x_vector_only_mode=False,
            )
-        
+
        # Run blocking operation in thread pool
        voice_prompt_items = await asyncio.to_thread(_create_prompt_sync)
-        
+
        # Cache if enabled
        if use_cache:
            cache_key = get_cache_key(audio_path, reference_text)
            cache_voice_prompt(cache_key, voice_prompt_items)
-        
+
        return voice_prompt_items, False
-    
+
    async def combine_voice_prompts(
        self,
        audio_paths: List[str],
        reference_texts: List[str],
    ) -> Tuple[np.ndarray, str]:
-        """
-        Combine multiple reference samples for better quality.
-        
-        Args:
-            audio_paths: List of audio file paths
-            reference_texts: List of reference texts
-            
-        Returns:
-            Tuple of (combined_audio, combined_text)
-        """
-        combined_audio = []
-        
-        for audio_path in audio_paths:
-            audio, sr = load_audio(audio_path)
-            audio = normalize_audio(audio)
-            combined_audio.append(audio)
-        
-        # Concatenate audio
-        mixed = np.concatenate(combined_audio)
-        mixed = normalize_audio(mixed)
-        
-        # Combine texts
-        combined_text = " ".join(reference_texts)
-        
-        return mixed, combined_text
-    
+        return await _combine_voice_prompts(audio_paths, reference_texts)
+
    async def generate(
        self,
        text: str,
@@ -376,15 +232,6 @@ class PyTorchTTSBackend:
        return audio, sample_rate


-WHISPER_HF_REPOS = {
-    "base": "openai/whisper-base",
-    "small": "openai/whisper-small",
-    "medium": "openai/whisper-medium",
-    "large": "openai/whisper-large-v3",
-    "turbo": "openai/whisper-large-v3-turbo",
-}
-
-
 class PyTorchSTTBackend:
    """PyTorch-based STT backend using Whisper."""

@@ -393,72 +240,18 @@ class PyTorchSTTBackend:
        self.processor = None
        self.model_size = model_size
        self.device = self._get_device()
-    
+
    def _get_device(self) -> str:
        """Get the best available device."""
-        if torch.cuda.is_available():
-            return "cuda"
-        # Intel Arc / Intel Xe GPU via intel-extension-for-pytorch (IPEX)
-        try:
-            import intel_extension_for_pytorch  # noqa: F401
-            if hasattr(torch, 'xpu') and torch.xpu.is_available():
-                return "xpu"
-        except ImportError:
-            pass
-        # Any GPU on Windows via DirectML (torch-directml)
-        try:
-            import torch_directml
-            if torch_directml.device_count() > 0:
-                return torch_directml.device(0)
-        except ImportError:
-            pass
-        if hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
-            return "cpu"  # MPS disabled for stability
-        return "cpu"
-    
+        return get_torch_device(allow_xpu=True, allow_directml=True)
+
    def is_loaded(self) -> bool:
        """Check if model is loaded."""
        return self.model is not None
-    
+
    def _is_model_cached(self, model_size: str) -> bool:
-        """
-        Check if the Whisper model is already cached locally AND fully downloaded.
-        
-        Args:
-            model_size: Model size to check
-            
-        Returns:
-            True if model is fully cached, False if missing or incomplete
-        """
-        try:
-            from huggingface_hub import constants as hf_constants
-            hf_repo = WHISPER_HF_REPOS.get(model_size, f"openai/whisper-{model_size}")
-            repo_cache = Path(hf_constants.HF_HUB_CACHE) / ("models--" + hf_repo.replace("/", "--"))
-
-            if not repo_cache.exists():
-                return False
-
-            # Check for .incomplete files - if any exist, download is still in progress
-            blobs_dir = repo_cache / "blobs"
-            if blobs_dir.exists() and any(blobs_dir.glob("*.incomplete")):
-                print(f"[_is_model_cached] Found .incomplete files for whisper-{model_size}, treating as not cached")
-                return False
-
-            # Check that actual model weight files exist in snapshots
-            snapshots_dir = repo_cache / "snapshots"
-            if snapshots_dir.exists():
-                has_weights = (
-                    any(snapshots_dir.rglob("*.safetensors")) or
-                    any(snapshots_dir.rglob("*.bin"))
-                )
-                if not has_weights:
-                    print(f"[_is_model_cached] No model weights found for whisper-{model_size}, treating as not cached")
-                    return False
-
-            return True
-        except Exception as e:
-            print(f"[_is_model_cached] Error checking cache for whisper-{model_size}: {e}")
-            return False
+        hf_repo = WHISPER_HF_REPOS.get(model_size, f"openai/whisper-{model_size}")
+        return is_model_cached(hf_repo)

    async def load_model_async(self, model_size: Optional[str] = None):
        """
@@ -467,95 +260,35 @@ class PyTorchSTTBackend:
        Args:
            model_size: Model size (tiny, base, small, medium, large)
        """
-        print(f"[DEBUG] load_model_async called with size: {model_size}")
        if model_size is None:
            model_size = self.model_size

-        print(f"[DEBUG] Model already loaded? {self.model is not None}, current size: {self.model_size}, requested: {model_size}")
        if self.model is not None and self.model_size == model_size:
-            print(f"[DEBUG] Early return - model already loaded")
            return

-        print(f"[DEBUG] Calling asyncio.to_thread for _load_model_sync")
-        # Run blocking load in thread pool
        await asyncio.to_thread(self._load_model_sync, model_size)
-        print(f"[DEBUG] asyncio.to_thread completed")
-    
+
    # Alias for compatibility
    load_model = load_model_async
-    
+
    def _load_model_sync(self, model_size: str):
        """Synchronous model loading."""
-        print(f"[DEBUG] _load_model_sync called for Whisper {model_size}")
-        try:
-            progress_manager = get_progress_manager()
-            task_manager = get_task_manager()
-            progress_model_name = f"whisper-{model_size}"
+        progress_model_name = f"whisper-{model_size}"
+        is_cached = self._is_model_cached(model_size)

-            # Check if model is already cached
-            is_cached = self._is_model_cached(model_size)
-
-            # Set up progress callback and tracker
-            # If cached: filter out non-download progress
-            # If not cached: report all progress (we're actually downloading)
-            progress_callback = create_hf_progress_callback(progress_model_name, progress_manager)
-            tracker = HFProgressTracker(progress_callback, filter_non_downloads=is_cached)
-
-            # Patch tqdm BEFORE importing transformers
-            print("[DEBUG] Starting tqdm patch BEFORE transformers import")
-            tracker_context = tracker.patch_download()
-            tracker_context.__enter__()
-            print("[DEBUG] tqdm patched, now importing transformers")
-
-            # Import transformers
+        with model_load_progress(progress_model_name, is_cached):
            from transformers import WhisperProcessor, WhisperForConditionalGeneration

            model_name = WHISPER_HF_REPOS.get(model_size, f"openai/whisper-{model_size}")
-            print(f"[DEBUG] Model name: {model_name}")
+            logger.info("Loading Whisper model %s on %s...", model_size, self.device)

-            print(f"Loading Whisper model {model_size} on {self.device}...")
+            self.processor = WhisperProcessor.from_pretrained(model_name)
+            self.model = WhisperForConditionalGeneration.from_pretrained(model_name)

-            # Only track download progress if model is NOT cached
-            if not is_cached:
-                # Start tracking download task
-                task_manager.start_download(progress_model_name)
+        self.model.to(self.device)
+        self.model_size = model_size
+        logger.info("Whisper model %s loaded successfully", model_size)

-                # Initialize progress state so SSE endpoint has initial data to send
-                progress_manager.update_progress(
-                    model_name=progress_model_name,
-                    current=0,
-                    total=0,  # Will be updated once actual total is known
-                    filename="Connecting to HuggingFace...",
-                    status="downloading",
-                )
-
-            # Load models (tqdm is patched, but filters out non-download progress)
-            try:
-                self.processor = WhisperProcessor.from_pretrained(model_name)
-                self.model = WhisperForConditionalGeneration.from_pretrained(model_name)
-            finally:
-                # Exit the patch context
-                tracker_context.__exit__(None, None, None)
-            
-            # Only mark download as complete if we were tracking it
-            if not is_cached:
-                progress_manager.mark_complete(progress_model_name)
-                task_manager.complete_download(progress_model_name)
-            
-            self.model.to(self.device)
-            self.model_size = model_size
-            
-            print(f"Whisper model {model_size} loaded successfully")
-            
-        except Exception as e:
-            print(f"Error loading Whisper model: {e}")
-            progress_manager = get_progress_manager()
-            task_manager = get_task_manager()
-            progress_model_name = f"whisper-{model_size}"
-            progress_manager.mark_error(progress_model_name, str(e))
-            task_manager.error_download(progress_model_name, str(e))
-            raise
-    
    def unload_model(self):
        """Unload the model to free memory."""
        if self.model is not None:
@@ -563,12 +296,12 @@ class PyTorchSTTBackend:
            del self.processor
            self.model = None
            self.processor = None
-            
+
            if torch.cuda.is_available():
                torch.cuda.empty_cache()
-            
-            print("Whisper model unloaded")
-    
+
+            logger.info("Whisper model unloaded")
+
    async def transcribe(
        self,
        audio_path: str,
@@ -576,21 +309,21 @@ class PyTorchSTTBackend:
    ) -> str:
        """
        Transcribe audio to text.
-        
+
        Args:
            audio_path: Path to audio file
            language: Optional language hint (en or zh)
-            
+
        Returns:
            Transcribed text
        """
        await self.load_model_async(None)
-        
+
        def _transcribe_sync():
            """Run synchronous transcription in thread pool."""
            # Load audio
            audio, sr = load_audio(audio_path, sample_rate=16000)
-            
+
            # Process audio
            inputs = self.processor(
                audio,
@@ -598,7 +331,7 @@ class PyTorchSTTBackend:
                return_tensors="pt",
            )
            inputs = inputs.to(self.device)
-            
+
            # Generate transcription
            # If language is provided, force it; otherwise let Whisper auto-detect
            generate_kwargs = {}
@@ -608,20 +341,20 @@ class PyTorchSTTBackend:
                    task="transcribe",
                )
                generate_kwargs["forced_decoder_ids"] = forced_decoder_ids
-            
+
            with torch.no_grad():
                predicted_ids = self.model.generate(
                    inputs["input_features"],
                    **generate_kwargs,
                )
-            
+
            # Decode
            transcription = self.processor.batch_decode(
                predicted_ids,
                skip_special_tokens=True,
            )[0]
-            
+
            return transcription.strip()
-        
+
        # Run blocking transcription in thread pool
        return await asyncio.to_thread(_transcribe_sync)
@@ -8,11 +8,14 @@ Usage:

 import PyInstaller.__main__
 import argparse
+import logging
 import os
 import platform
 import sys
 from pathlib import Path

+logger = logging.getLogger(__name__)
+

 def is_apple_silicon():
    """Check if running on Apple Silicon."""
@@ -28,125 +31,246 @@ def build_server(cuda=False):
    """
    backend_dir = Path(__file__).parent

-    binary_name = 'voicebox-server-cuda' if cuda else 'voicebox-server'
+    binary_name = "voicebox-server-cuda" if cuda else "voicebox-server"

    # PyInstaller arguments
    args = [
-        'server.py',  # Use server.py as entry point instead of main.py
-        '--onefile',
-        '--noconsole',  # No visible console window on Windows
-        '--name', binary_name,
+        "server.py",  # Use server.py as entry point instead of main.py
+        "--onefile",
+        "--name",
+        binary_name,
    ]

+    # Hide console window on Windows only. On macOS/Linux the sidecar needs
+    # stdout/stderr for Tauri to capture logs.
+    if platform.system() == "Windows":
+        args.append("--noconsole")
+
    # Add local qwen_tts path if specified (for editable installs)
-    qwen_tts_path = os.getenv('QWEN_TTS_PATH')
+    qwen_tts_path = os.getenv("QWEN_TTS_PATH")
    if qwen_tts_path and Path(qwen_tts_path).exists():
-        args.extend(['--paths', str(qwen_tts_path)])
-        print(f"Using local qwen_tts source from: {qwen_tts_path}")
+        args.extend(["--paths", str(qwen_tts_path)])
+        logger.info("Using local qwen_tts source from: %s", qwen_tts_path)

    # Add common hidden imports
-    args.extend([
-        '--hidden-import', 'backend',
-        '--hidden-import', 'backend.main',
-        '--hidden-import', 'backend.config',
-        '--hidden-import', 'backend.database',
-        '--hidden-import', 'backend.models',
-        '--hidden-import', 'backend.profiles',
-        '--hidden-import', 'backend.history',
-        '--hidden-import', 'backend.tts',
-        '--hidden-import', 'backend.transcribe',
-        '--hidden-import', 'backend.platform_detect',
-        '--hidden-import', 'backend.backends',
-        '--hidden-import', 'backend.backends.pytorch_backend',
-        '--hidden-import', 'backend.utils.audio',
-        '--hidden-import', 'backend.utils.cache',
-        '--hidden-import', 'backend.utils.progress',
-        '--hidden-import', 'backend.utils.hf_progress',
-        '--hidden-import', 'backend.utils.validation',
-        '--hidden-import', 'backend.cuda_download',
-        '--hidden-import', 'backend.effects',
-        '--hidden-import', 'backend.utils.effects',
-        '--hidden-import', 'backend.versions',
-        '--hidden-import', 'pedalboard',
-        '--hidden-import', 'torch',
-        '--hidden-import', 'transformers',
-        '--hidden-import', 'fastapi',
-        '--hidden-import', 'uvicorn',
-        '--hidden-import', 'sqlalchemy',
-        '--hidden-import', 'librosa',
-        '--hidden-import', 'soundfile',
-        '--hidden-import', 'qwen_tts',
-        '--hidden-import', 'qwen_tts.inference',
-        '--hidden-import', 'qwen_tts.inference.qwen3_tts_model',
-        '--hidden-import', 'qwen_tts.inference.qwen3_tts_tokenizer',
-        '--hidden-import', 'qwen_tts.core',
-        '--hidden-import', 'qwen_tts.cli',
-        '--copy-metadata', 'qwen-tts',
-        '--collect-submodules', 'qwen_tts',
-        '--collect-data', 'qwen_tts',
-        # Fix for pkg_resources and jaraco namespace packages
-        '--hidden-import', 'pkg_resources.extern',
-        '--collect-submodules', 'jaraco',
-    ])
+    args.extend(
+        [
+            "--hidden-import",
+            "backend",
+            "--hidden-import",
+            "backend.main",
+            "--hidden-import",
+            "backend.config",
+            "--hidden-import",
+            "backend.database",
+            "--hidden-import",
+            "backend.models",
+            "--hidden-import",
+            "backend.services.profiles",
+            "--hidden-import",
+            "backend.services.history",
+            "--hidden-import",
+            "backend.services.tts",
+            "--hidden-import",
+            "backend.services.transcribe",
+            "--hidden-import",
+            "backend.utils.platform_detect",
+            "--hidden-import",
+            "backend.backends",
+            "--hidden-import",
+            "backend.backends.pytorch_backend",
+            "--hidden-import",
+            "backend.utils.audio",
+            "--hidden-import",
+            "backend.utils.cache",
+            "--hidden-import",
+            "backend.utils.progress",
+            "--hidden-import",
+            "backend.utils.hf_progress",
+            "--hidden-import",
+            "backend.services.cuda",
+            "--hidden-import",
+            "backend.services.effects",
+            "--hidden-import",
+            "backend.utils.effects",
+            "--hidden-import",
+            "backend.services.versions",
+            "--hidden-import",
+            "pedalboard",
+            "--hidden-import",
+            "chatterbox",
+            "--hidden-import",
+            "chatterbox.tts_turbo",
+            "--hidden-import",
+            "chatterbox.mtl_tts",
+            "--hidden-import",
+            "backend.backends.chatterbox_backend",
+            "--hidden-import",
+            "backend.backends.chatterbox_turbo_backend",
+            "--hidden-import",
+            "backend.backends.luxtts_backend",
+            "--hidden-import",
+            "zipvoice",
+            "--hidden-import",
+            "zipvoice.luxvoice",
+            "--collect-all",
+            "zipvoice",
+            "--collect-all",
+            "linacodec",
+            "--hidden-import",
+            "torch",
+            "--hidden-import",
+            "transformers",
+            "--hidden-import",
+            "fastapi",
+            "--hidden-import",
+            "uvicorn",
+            "--hidden-import",
+            "sqlalchemy",
+            "--hidden-import",
+            "librosa",
+            "--hidden-import",
+            "soundfile",
+            "--hidden-import",
+            "qwen_tts",
+            "--hidden-import",
+            "qwen_tts.inference",
+            "--hidden-import",
+            "qwen_tts.inference.qwen3_tts_model",
+            "--hidden-import",
+            "qwen_tts.inference.qwen3_tts_tokenizer",
+            "--hidden-import",
+            "qwen_tts.core",
+            "--hidden-import",
+            "qwen_tts.cli",
+            "--copy-metadata",
+            "qwen-tts",
+            "--copy-metadata",
+            "requests",
+            "--copy-metadata",
+            "transformers",
+            "--copy-metadata",
+            "huggingface-hub",
+            "--copy-metadata",
+            "tokenizers",
+            "--copy-metadata",
+            "safetensors",
+            "--copy-metadata",
+            "tqdm",
+            "--hidden-import",
+            "requests",
+            "--collect-submodules",
+            "qwen_tts",
+            "--collect-data",
+            "qwen_tts",
+            # Fix for pkg_resources and jaraco namespace packages
+            "--hidden-import",
+            "pkg_resources.extern",
+            "--collect-submodules",
+            "jaraco",
+            # inflect uses typeguard @typechecked which calls inspect.getsource()
+            # at import time — needs .py source files, not just .pyc bytecode
+            "--collect-all",
+            "inflect",
+            # perth ships pretrained watermark model files (hparams.yaml, .pth.tar)
+            # in perth/perth_net/pretrained/ — needed by chatterbox at runtime
+            "--collect-all",
+            "perth",
+            # piper_phonemize ships espeak-ng-data/ (phoneme tables, language dicts)
+            # needed by LuxTTS for text-to-phoneme conversion
+            "--collect-all",
+            "piper_phonemize",
+        ]
+    )

    # Add CUDA-specific hidden imports
    if cuda:
-        print("Building with CUDA support")
-        args.extend([
-            '--hidden-import', 'torch.cuda',
-            '--hidden-import', 'torch.backends.cudnn',
-        ])
+        logger.info("Building with CUDA support")
+        args.extend(
+            [
+                "--hidden-import",
+                "torch.cuda",
+                "--hidden-import",
+                "torch.backends.cudnn",
+            ]
+        )
    else:
        # Exclude NVIDIA CUDA packages from CPU-only builds to keep binary small.
        # When building from a venv with CUDA torch installed, PyInstaller would
        # bundle ~3GB of NVIDIA shared libraries. We exclude both the Python
        # modules and the binary DLLs.
        nvidia_packages = [
-            'nvidia', 'nvidia.cublas', 'nvidia.cuda_cupti', 'nvidia.cuda_nvrtc',
-            'nvidia.cuda_runtime', 'nvidia.cudnn', 'nvidia.cufft', 'nvidia.curand',
-            'nvidia.cusolver', 'nvidia.cusparse', 'nvidia.nccl', 'nvidia.nvjitlink',
-            'nvidia.nvtx',
+            "nvidia",
+            "nvidia.cublas",
+            "nvidia.cuda_cupti",
+            "nvidia.cuda_nvrtc",
+            "nvidia.cuda_runtime",
+            "nvidia.cudnn",
+            "nvidia.cufft",
+            "nvidia.curand",
+            "nvidia.cusolver",
+            "nvidia.cusparse",
+            "nvidia.nccl",
+            "nvidia.nvjitlink",
+            "nvidia.nvtx",
        ]
        for pkg in nvidia_packages:
-            args.extend(['--exclude-module', pkg])
+            args.extend(["--exclude-module", pkg])

    # Add MLX-specific imports if building on Apple Silicon (never for CUDA builds)
    if is_apple_silicon() and not cuda:
-        print("Building for Apple Silicon - including MLX dependencies")
-        args.extend([
-            '--hidden-import', 'backend.backends.mlx_backend',
-            '--hidden-import', 'mlx',
-            '--hidden-import', 'mlx.core',
-            '--hidden-import', 'mlx.nn',
-            '--hidden-import', 'mlx_audio',
-            '--hidden-import', 'mlx_audio.tts',
-            '--hidden-import', 'mlx_audio.stt',
-            '--collect-submodules', 'mlx',
-            '--collect-submodules', 'mlx_audio',
-            # Use --collect-all so PyInstaller bundles both data files AND
-            # native shared libraries (.dylib, .metallib) for MLX.
-            # Previously only --collect-data was used, which caused MLX to
-            # raise OSError at runtime inside the bundled binary because
-            # the Metal shader libraries were missing.
-            '--collect-all', 'mlx',
-            '--collect-all', 'mlx_audio',
-        ])
+        logger.info("Building for Apple Silicon - including MLX dependencies")
+        args.extend(
+            [
+                "--hidden-import",
+                "backend.backends.mlx_backend",
+                "--hidden-import",
+                "mlx",
+                "--hidden-import",
+                "mlx.core",
+                "--hidden-import",
+                "mlx.nn",
+                "--hidden-import",
+                "mlx_audio",
+                "--hidden-import",
+                "mlx_audio.tts",
+                "--hidden-import",
+                "mlx_audio.stt",
+                "--collect-submodules",
+                "mlx",
+                "--collect-submodules",
+                "mlx_audio",
+                # Use --collect-all so PyInstaller bundles both data files AND
+                # native shared libraries (.dylib, .metallib) for MLX.
+                # Previously only --collect-data was used, which caused MLX to
+                # raise OSError at runtime inside the bundled binary because
+                # the Metal shader libraries were missing.
+                "--collect-all",
+                "mlx",
+                "--collect-all",
+                "mlx_audio",
+            ]
+        )
    elif not cuda:
-        print("Building for non-Apple Silicon platform - PyTorch only")
+        logger.info("Building for non-Apple Silicon platform - PyTorch only")

-    dist_dir = str(backend_dir / 'dist')
-    build_dir = str(backend_dir / 'build')
+    dist_dir = str(backend_dir / "dist")
+    build_dir = str(backend_dir / "build")

-    args.extend([
-        '--distpath', dist_dir,
-        '--workpath', build_dir,
-        '--noconfirm',
-        '--clean',
-    ])
+    args.extend(
+        [
+            "--distpath",
+            dist_dir,
+            "--workpath",
+            build_dir,
+            "--noconfirm",
+            "--clean",
+        ]
+    )

    # Change to backend directory
    os.chdir(backend_dir)
-    
+
    # For CPU builds on Windows, ensure we're using CPU-only torch.
    # If CUDA torch is installed (local dev), swap to CPU torch before building,
    # then restore CUDA torch after. This prevents PyInstaller from bundling
@@ -154,17 +278,28 @@ def build_server(cuda=False):
    restore_cuda = False
    if not cuda and platform.system() == "Windows":
        import subprocess
+
        result = subprocess.run(
-            [sys.executable, "-c", "import torch; print(torch.version.cuda or '')"],
-            capture_output=True, text=True
+            [sys.executable, "-c", "import torch; print(torch.version.cuda or '')"], capture_output=True, text=True
        )
        has_cuda_torch = bool(result.stdout.strip())
        if has_cuda_torch:
-            print("CUDA torch detected — installing CPU torch for CPU build...")
+            logger.info("CUDA torch detected — installing CPU torch for CPU build...")
            subprocess.run(
-                [sys.executable, "-m", "pip", "install", "torch", "torchvision", "torchaudio",
-                 "--index-url", "https://download.pytorch.org/whl/cpu", "--force-reinstall", "-q"],
-                check=True
+                [
+                    sys.executable,
+                    "-m",
+                    "pip",
+                    "install",
+                    "torch",
+                    "torchvision",
+                    "torchaudio",
+                    "--index-url",
+                    "https://download.pytorch.org/whl/cpu",
+                    "--force-reinstall",
+                    "-q",
+                ],
+                check=True,
            )
            restore_cuda = True

@@ -174,57 +309,34 @@ def build_server(cuda=False):
    finally:
        # Restore CUDA torch if we swapped it out (even on build failure)
        if restore_cuda:
-            print("Restoring CUDA torch...")
+            logger.info("Restoring CUDA torch...")
            import subprocess
+
            subprocess.run(
-                [sys.executable, "-m", "pip", "install", "torch", "torchvision", "torchaudio",
-                 "--index-url", "https://download.pytorch.org/whl/cu126", "--force-reinstall", "-q"],
-                check=True
+                [
+                    sys.executable,
+                    "-m",
+                    "pip",
+                    "install",
+                    "torch",
+                    "torchvision",
+                    "torchaudio",
+                    "--index-url",
+                    "https://download.pytorch.org/whl/cu126",
+                    "--force-reinstall",
+                    "-q",
+                ],
+                check=True,
            )
-    
-    print(f"Binary built in {backend_dir / 'dist' / binary_name}")
+
+    logger.info("Binary built in %s", backend_dir / "dist" / binary_name)


-def _get_cuda_dll_excludes():
-    """Get list of CUDA DLL filenames to exclude from CPU builds.
-    
-    When building locally with CUDA torch installed, PyInstaller bundles ~3GB of
-    CUDA DLLs from torch/lib/. Returns a list of DLL filenames to exclude.
-    """
-    try:
-        import torch
-        torch_lib = Path(torch.__file__).parent / 'lib'
-    except ImportError:
-        return []
-    
-    cuda_prefixes = (
-        'torch_cuda', 'cublas', 'cublasLt', 'cudnn', 'cusparse', 'cufft',
-        'cusolver', 'cusolverMg', 'curand', 'nvrtc', 'nvJitLink', 'nccl',
-        'nvperf', 'nvrtc-builtins',
-    )
-    
-    exclude_dlls = []
-    if torch_lib.exists():
-        for f in torch_lib.iterdir():
-            if f.suffix == '.dll' and any(f.name.startswith(p) for p in cuda_prefixes):
-                exclude_dlls.append(f.name)
-    
-    if exclude_dlls:
-        total_mb = sum(
-            (torch_lib / dll).stat().st_size 
-            for dll in exclude_dlls 
-            if (torch_lib / dll).exists()
-        ) / 1024 / 1024
-        print(f"CPU build: will exclude {len(exclude_dlls)} CUDA DLLs ({total_mb:.0f} MB)")
-    
-    return exclude_dlls
-
-
-if __name__ == '__main__':
+if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="Build voicebox-server binary")
    parser.add_argument(
-        '--cuda',
-        action='store_true',
+        "--cuda",
+        action="store_true",
        help="Build CUDA-enabled binary (voicebox-server-cuda)",
    )
    cli_args = parser.parse_args()
@@ -4,20 +4,24 @@ Configuration module for voicebox backend.
 Handles data directory configuration for production bundling.
 """

+import logging
 import os
 from pathlib import Path

+logger = logging.getLogger(__name__)
+
 # Allow users to override the HuggingFace model download directory.
 # Set VOICEBOX_MODELS_DIR to an absolute path before starting the server.
 # This sets HF_HUB_CACHE so all huggingface_hub downloads go to that path.
 _custom_models_dir = os.environ.get("VOICEBOX_MODELS_DIR")
 if _custom_models_dir:
    os.environ["HF_HUB_CACHE"] = _custom_models_dir
-    print(f"[config] Model download path set to: {_custom_models_dir}")
+    logger.info("Model download path set to: %s", _custom_models_dir)

 # Default data directory (used in development)
 _data_dir = Path("data")

+
 def set_data_dir(path: str | Path):
    """
    Set the data directory path.
@@ -28,7 +32,8 @@ def set_data_dir(path: str | Path):
    global _data_dir
    _data_dir = Path(path)
    _data_dir.mkdir(parents=True, exist_ok=True)
-    print(f"Data directory set to: {_data_dir.absolute()}")
+    logger.info("Data directory set to: %s", _data_dir.absolute())
+

 def get_data_dir() -> Path:
    """
@@ -39,28 +44,33 @@ def get_data_dir() -> Path:
    """
    return _data_dir

+
 def get_db_path() -> Path:
    """Get database file path."""
    return _data_dir / "voicebox.db"

+
 def get_profiles_dir() -> Path:
    """Get profiles directory path."""
    path = _data_dir / "profiles"
    path.mkdir(parents=True, exist_ok=True)
    return path

+
 def get_generations_dir() -> Path:
    """Get generations directory path."""
    path = _data_dir / "generations"
    path.mkdir(parents=True, exist_ok=True)
    return path

+
 def get_cache_dir() -> Path:
    """Get cache directory path."""
    path = _data_dir / "cache"
    path.mkdir(parents=True, exist_ok=True)
    return path

+
 def get_models_dir() -> Path:
    """Get models directory path."""
    path = _data_dir / "models"
@@ -1,487 +0,0 @@
-"""
-SQLite database ORM using SQLAlchemy.
-"""
-
-from sqlalchemy import create_engine, Column, String, Integer, Float, DateTime, Text, ForeignKey, Boolean
-from sqlalchemy.ext.declarative import declarative_base
-from sqlalchemy.orm import sessionmaker, Session
-from datetime import datetime
-import uuid
-from pathlib import Path
-
-from . import config
-
-Base = declarative_base()
-
-
-class VoiceProfile(Base):
-    """Voice profile database model."""
-    __tablename__ = "profiles"
-
-    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
-    name = Column(String, unique=True, nullable=False)
-    description = Column(Text)
-    language = Column(String, default="en")
-    avatar_path = Column(String, nullable=True)
-    effects_chain = Column(Text, nullable=True)  # JSON-serialized default effects chain
-    created_at = Column(DateTime, default=datetime.utcnow)
-    updated_at = Column(DateTime, default=datetime.utcnow, onupdate=datetime.utcnow)
-
-
-class ProfileSample(Base):
-    """Voice profile sample database model."""
-    __tablename__ = "profile_samples"
-    
-    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
-    profile_id = Column(String, ForeignKey("profiles.id"), nullable=False)
-    audio_path = Column(String, nullable=False)
-    reference_text = Column(Text, nullable=False)
-
-
-class Generation(Base):
-    """Generation history database model."""
-    __tablename__ = "generations"
-
-    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
-    profile_id = Column(String, ForeignKey("profiles.id"), nullable=False)
-    text = Column(Text, nullable=False)
-    language = Column(String, default="en")
-    audio_path = Column(String, nullable=True)
-    duration = Column(Float, nullable=True)
-    seed = Column(Integer)
-    instruct = Column(Text)
-    engine = Column(String, default="qwen")
-    model_size = Column(String, nullable=True)
-    status = Column(String, default="completed")  # generating, completed, failed
-    error = Column(Text, nullable=True)
-    is_favorited = Column(Boolean, default=False)
-    created_at = Column(DateTime, default=datetime.utcnow)
-
-
-class Story(Base):
-    """Story database model."""
-    __tablename__ = "stories"
-    
-    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
-    name = Column(String, nullable=False)
-    description = Column(Text)
-    created_at = Column(DateTime, default=datetime.utcnow)
-    updated_at = Column(DateTime, default=datetime.utcnow, onupdate=datetime.utcnow)
-
-
-class StoryItem(Base):
-    """Story item database model (links generations to stories)."""
-    __tablename__ = "story_items"
-    
-    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
-    story_id = Column(String, ForeignKey("stories.id"), nullable=False)
-    generation_id = Column(String, ForeignKey("generations.id"), nullable=False)
-    version_id = Column(String, ForeignKey("generation_versions.id"), nullable=True)  # Pin to specific version, null = use generation default
-    start_time_ms = Column(Integer, nullable=False, default=0)  # Milliseconds from story start
-    track = Column(Integer, nullable=False, default=0)  # Track number (0 = main track)
-    trim_start_ms = Column(Integer, nullable=False, default=0)  # Milliseconds trimmed from start
-    trim_end_ms = Column(Integer, nullable=False, default=0)  # Milliseconds trimmed from end
-    created_at = Column(DateTime, default=datetime.utcnow)
-
-
-class Project(Base):
-    """Audio studio project database model."""
-    __tablename__ = "projects"
-    
-    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
-    name = Column(String, nullable=False)
-    data = Column(Text)  # JSON string
-    created_at = Column(DateTime, default=datetime.utcnow)
-    updated_at = Column(DateTime, default=datetime.utcnow, onupdate=datetime.utcnow)
-
-
-class GenerationVersion(Base):
-    """A version of a generation's audio (clean, processed, alternate takes)."""
-    __tablename__ = "generation_versions"
-
-    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
-    generation_id = Column(String, ForeignKey("generations.id"), nullable=False)
-    label = Column(String, nullable=False)  # "clean", "processed", or user-defined
-    audio_path = Column(String, nullable=False)
-    effects_chain = Column(Text, nullable=True)  # JSON-serialized effects config, null for clean
-    source_version_id = Column(String, ForeignKey("generation_versions.id"), nullable=True)  # Which version was used as input
-    is_default = Column(Boolean, default=False)
-    created_at = Column(DateTime, default=datetime.utcnow)
-
-
-class EffectPreset(Base):
-    """Saved effect chain preset."""
-    __tablename__ = "effect_presets"
-
-    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
-    name = Column(String, unique=True, nullable=False)
-    description = Column(Text, nullable=True)
-    effects_chain = Column(Text, nullable=False)  # JSON-serialized effects config
-    is_builtin = Column(Boolean, default=False)
-    sort_order = Column(Integer, default=100)
-    created_at = Column(DateTime, default=datetime.utcnow)
-
-
-class AudioChannel(Base):
-    """Audio channel (bus) database model."""
-    __tablename__ = "audio_channels"
-    
-    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
-    name = Column(String, nullable=False)
-    is_default = Column(Boolean, default=False)
-    created_at = Column(DateTime, default=datetime.utcnow)
-
-
-class ChannelDeviceMapping(Base):
-    """Mapping between channels and OS audio devices."""
-    __tablename__ = "channel_device_mappings"
-    
-    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
-    channel_id = Column(String, ForeignKey("audio_channels.id"), nullable=False)
-    device_id = Column(String, nullable=False)  # OS device identifier
-
-
-class ProfileChannelMapping(Base):
-    """Mapping between voice profiles and audio channels (many-to-many)."""
-    __tablename__ = "profile_channel_mappings"
-    
-    profile_id = Column(String, ForeignKey("profiles.id"), primary_key=True)
-    channel_id = Column(String, ForeignKey("audio_channels.id"), primary_key=True)
-
-
-# Database setup will be initialized in init_db()
-engine = None
-SessionLocal = None
-_db_path = None
-
-
-def init_db():
-    """Initialize database tables."""
-    global engine, SessionLocal, _db_path
-
-    _db_path = config.get_db_path()
-    _db_path.parent.mkdir(parents=True, exist_ok=True)
-
-    engine = create_engine(
-        f"sqlite:///{_db_path}",
-        connect_args={"check_same_thread": False},
-    )
-
-    SessionLocal = sessionmaker(autocommit=False, autoflush=False, bind=engine)
-    
-    # Run migrations before creating tables
-    _run_migrations(engine)
-    
-    Base.metadata.create_all(bind=engine)
-    
-    # Create default channel if it doesn't exist
-    db = SessionLocal()
-    try:
-        default_channel = db.query(AudioChannel).filter(AudioChannel.is_default == True).first()
-        if not default_channel:
-            default_channel = AudioChannel(
-                id=str(uuid.uuid4()),
-                name="Default",
-                is_default=True
-            )
-            db.add(default_channel)
-            
-            # Assign all existing profiles to default channel
-            profiles = db.query(VoiceProfile).all()
-            for profile in profiles:
-                mapping = ProfileChannelMapping(
-                    profile_id=profile.id,
-                    channel_id=default_channel.id
-                )
-                db.add(mapping)
-            
-            db.commit()
-    finally:
-        db.close()
-
-    # Backfill: create "clean" GenerationVersion entries for existing generations
-    _backfill_generation_versions()
-
-    # Seed built-in effect presets
-    _seed_builtin_presets()
-
-
-def _run_migrations(engine):
-    """Run database migrations."""
-    from sqlalchemy import inspect, text
-    
-    inspector = inspect(engine)
-    
-    # Check if story_items table exists
-    if 'story_items' not in inspector.get_table_names():
-        return  # Table doesn't exist yet, will be created fresh
-    
-    # Get columns in story_items table
-    columns = {col['name'] for col in inspector.get_columns('story_items')}
-    
-    # Migration: Remove position column and ensure start_time_ms exists
-    # SQLite doesn't support DROP COLUMN easily, so we recreate the table
-    if 'position' in columns:
-        print("Migrating story_items: removing position column, using start_time_ms")
-        
-        with engine.connect() as conn:
-            # Check if start_time_ms already exists
-            has_start_time = 'start_time_ms' in columns
-            
-            if not has_start_time:
-                # First, add the new column temporarily
-                conn.execute(text("ALTER TABLE story_items ADD COLUMN start_time_ms INTEGER DEFAULT 0"))
-                
-                # Calculate timecodes from position ordering
-                result = conn.execute(text("""
-                    SELECT si.id, si.story_id, si.position, g.duration
-                    FROM story_items si
-                    JOIN generations g ON si.generation_id = g.id
-                    ORDER BY si.story_id, si.position
-                """))
-                
-                rows = result.fetchall()
-                
-                current_story_id = None
-                current_time_ms = 0
-                
-                for row in rows:
-                    item_id, story_id, position, duration = row
-                    
-                    if story_id != current_story_id:
-                        current_story_id = story_id
-                        current_time_ms = 0
-                    
-                    conn.execute(
-                        text("UPDATE story_items SET start_time_ms = :time WHERE id = :id"),
-                        {"time": current_time_ms, "id": item_id}
-                    )
-                    
-                    current_time_ms += int(duration * 1000) + 200
-                
-                conn.commit()
-            
-            # Now recreate the table without the position column
-            # 1. Create new table
-            conn.execute(text("""
-                CREATE TABLE story_items_new (
-                    id VARCHAR PRIMARY KEY,
-                    story_id VARCHAR NOT NULL,
-                    generation_id VARCHAR NOT NULL,
-                    start_time_ms INTEGER NOT NULL DEFAULT 0,
-                    created_at DATETIME,
-                    FOREIGN KEY (story_id) REFERENCES stories(id),
-                    FOREIGN KEY (generation_id) REFERENCES generations(id)
-                )
-            """))
-            
-            # 2. Copy data
-            conn.execute(text("""
-                INSERT INTO story_items_new (id, story_id, generation_id, start_time_ms, created_at)
-                SELECT id, story_id, generation_id, start_time_ms, created_at FROM story_items
-            """))
-            
-            # 3. Drop old table
-            conn.execute(text("DROP TABLE story_items"))
-            
-            # 4. Rename new table
-            conn.execute(text("ALTER TABLE story_items_new RENAME TO story_items"))
-            
-            conn.commit()
-            print("Migrated story_items table to use start_time_ms (removed position column)")
-    
-    # Migration: Add track column if it doesn't exist
-    # Re-check columns after potential position migration
-    columns = {col['name'] for col in inspector.get_columns('story_items')}
-    if 'track' not in columns:
-        print("Migrating story_items: adding track column")
-        with engine.connect() as conn:
-            conn.execute(text("ALTER TABLE story_items ADD COLUMN track INTEGER NOT NULL DEFAULT 0"))
-            conn.commit()
-            print("Added track column to story_items")
-    
-    # Migration: Add trim columns if they don't exist
-    # Re-check columns after potential track migration
-    columns = {col['name'] for col in inspector.get_columns('story_items')}
-    if 'trim_start_ms' not in columns:
-        print("Migrating story_items: adding trim_start_ms column")
-        with engine.connect() as conn:
-            conn.execute(text("ALTER TABLE story_items ADD COLUMN trim_start_ms INTEGER NOT NULL DEFAULT 0"))
-            conn.commit()
-            print("Added trim_start_ms column to story_items")
-    
-    columns = {col['name'] for col in inspector.get_columns('story_items')}
-    if 'trim_end_ms' not in columns:
-        print("Migrating story_items: adding trim_end_ms column")
-        with engine.connect() as conn:
-            conn.execute(text("ALTER TABLE story_items ADD COLUMN trim_end_ms INTEGER NOT NULL DEFAULT 0"))
-            conn.commit()
-            print("Added trim_end_ms column to story_items")
-
-    # Migration: Add avatar_path to profiles table
-    if 'profiles' in inspector.get_table_names():
-        columns = {col['name'] for col in inspector.get_columns('profiles')}
-        if 'avatar_path' not in columns:
-            print("Migrating profiles: adding avatar_path column")
-            with engine.connect() as conn:
-                conn.execute(text("ALTER TABLE profiles ADD COLUMN avatar_path VARCHAR"))
-                conn.commit()
-                print("Added avatar_path column to profiles")
-
-    # Migration: Add status and error columns to generations table
-    if 'generations' in inspector.get_table_names():
-        columns = {col['name'] for col in inspector.get_columns('generations')}
-        if 'status' not in columns:
-            print("Migrating generations: adding status column")
-            with engine.connect() as conn:
-                conn.execute(text("ALTER TABLE generations ADD COLUMN status VARCHAR DEFAULT 'completed'"))
-                conn.commit()
-                print("Added status column to generations")
-        if 'error' not in columns:
-            print("Migrating generations: adding error column")
-            with engine.connect() as conn:
-                conn.execute(text("ALTER TABLE generations ADD COLUMN error TEXT"))
-                conn.commit()
-                print("Added error column to generations")
-        if 'engine' not in columns:
-            print("Migrating generations: adding engine column")
-            with engine.connect() as conn:
-                conn.execute(text("ALTER TABLE generations ADD COLUMN engine VARCHAR DEFAULT 'qwen'"))
-                conn.commit()
-                print("Added engine column to generations")
-        # Re-read columns after engine migration (variable name shadows outer `engine`)
-        columns = {col['name'] for col in inspector.get_columns('generations')}
-        if 'model_size' not in columns:
-            print("Migrating generations: adding model_size column")
-            with engine.connect() as conn:
-                conn.execute(text("ALTER TABLE generations ADD COLUMN model_size VARCHAR"))
-                conn.commit()
-                print("Added model_size column to generations")
-
-    # Migration: Add effects_chain to profiles table
-    if 'profiles' in inspector.get_table_names():
-        columns = {col['name'] for col in inspector.get_columns('profiles')}
-        if 'effects_chain' not in columns:
-            print("Migrating profiles: adding effects_chain column")
-            with engine.connect() as conn:
-                conn.execute(text("ALTER TABLE profiles ADD COLUMN effects_chain TEXT"))
-                conn.commit()
-                print("Added effects_chain column to profiles")
-
-    # Migration: Add sort_order to effect_presets table
-    if 'effect_presets' in inspector.get_table_names():
-        columns = {col['name'] for col in inspector.get_columns('effect_presets')}
-        if 'sort_order' not in columns:
-            print("Migrating effect_presets: adding sort_order column")
-            with engine.connect() as conn:
-                conn.execute(text("ALTER TABLE effect_presets ADD COLUMN sort_order INTEGER DEFAULT 100"))
-                conn.commit()
-                print("Added sort_order column to effect_presets")
-
-    # Migration: Add version_id column to story_items table
-    if 'story_items' in inspector.get_table_names():
-        columns = {col['name'] for col in inspector.get_columns('story_items')}
-        if 'version_id' not in columns:
-            print("Migrating story_items: adding version_id column")
-            with engine.connect() as conn:
-                conn.execute(text("ALTER TABLE story_items ADD COLUMN version_id VARCHAR"))
-                conn.commit()
-                print("Added version_id column to story_items")
-
-    # Migration: Add source_version_id to generation_versions table
-    if 'generation_versions' in inspector.get_table_names():
-        columns = {col['name'] for col in inspector.get_columns('generation_versions')}
-        if 'source_version_id' not in columns:
-            print("Migrating generation_versions: adding source_version_id column")
-            with engine.connect() as conn:
-                conn.execute(text("ALTER TABLE generation_versions ADD COLUMN source_version_id VARCHAR"))
-                conn.commit()
-                print("Added source_version_id column to generation_versions")
-
-    if 'generations' in inspector.get_table_names():
-        columns = {col['name'] for col in inspector.get_columns('generations')}
-        if 'is_favorited' not in columns:
-            print("Migrating generations: adding is_favorited column")
-            with engine.connect() as conn:
-                conn.execute(text("ALTER TABLE generations ADD COLUMN is_favorited BOOLEAN DEFAULT 0"))
-                conn.commit()
-                print("Added is_favorited column to generations")
-
-    # Migration: Create generation_versions for existing generations
-    # (populate after tables are created, handled in init_db)
-
-
-def _backfill_generation_versions():
-    """Create 'clean' version entries for existing generations that don't have any."""
-    db = SessionLocal()
-    try:
-        from pathlib import Path as _Path
-
-        # Find generations that have no version entries
-        existing_version_gen_ids = {
-            row[0] for row in db.query(GenerationVersion.generation_id).all()
-        }
-        generations = db.query(Generation).filter(
-            Generation.status == "completed",
-            Generation.audio_path.isnot(None),
-            Generation.audio_path != "",
-        ).all()
-
-        count = 0
-        for gen in generations:
-            if gen.id in existing_version_gen_ids:
-                continue
-            if not _Path(gen.audio_path).exists():
-                continue
-            version = GenerationVersion(
-                id=str(uuid.uuid4()),
-                generation_id=gen.id,
-                label="clean",
-                audio_path=gen.audio_path,
-                effects_chain=None,
-                is_default=True,
-            )
-            db.add(version)
-            count += 1
-
-        if count > 0:
-            db.commit()
-            print(f"Backfilled {count} generation version entries")
-    finally:
-        db.close()
-
-
-def _seed_builtin_presets():
-    """Ensure built-in effect presets exist in the database."""
-    import json
-    from .utils.effects import BUILTIN_PRESETS
-
-    db = SessionLocal()
-    try:
-        for idx, (key, preset_data) in enumerate(BUILTIN_PRESETS.items()):
-            sort_order = preset_data.get("sort_order", idx)
-            existing = db.query(EffectPreset).filter_by(name=preset_data["name"]).first()
-            if not existing:
-                preset = EffectPreset(
-                    id=str(uuid.uuid4()),
-                    name=preset_data["name"],
-                    description=preset_data.get("description"),
-                    effects_chain=json.dumps(preset_data["effects_chain"]),
-                    is_builtin=True,
-                    sort_order=sort_order,
-                )
-                db.add(preset)
-            elif existing.sort_order != sort_order:
-                existing.sort_order = sort_order
-        db.commit()
-    finally:
-        db.close()
-
-
-def get_db():
-    """Get database session (generator for dependency injection)."""
-    db = SessionLocal()
-    try:
-        yield db
-    finally:
-        db.close()
@@ -0,0 +1,44 @@
+"""Database package — ORM models, session management, and migrations.
+
+Re-exports all public symbols so that ``from .database import get_db``
+and ``from .database import Generation as DBGeneration`` continue to work
+without changing any importers.
+"""
+
+from .models import (
+    Base,
+    AudioChannel,
+    ChannelDeviceMapping,
+    EffectPreset,
+    Generation,
+    GenerationVersion,
+    ProfileChannelMapping,
+    ProfileSample,
+    Project,
+    Story,
+    StoryItem,
+    VoiceProfile,
+)
+from .session import engine, SessionLocal, _db_path, init_db, get_db
+
+__all__ = [
+    # Models
+    "Base",
+    "AudioChannel",
+    "ChannelDeviceMapping",
+    "EffectPreset",
+    "Generation",
+    "GenerationVersion",
+    "ProfileChannelMapping",
+    "ProfileSample",
+    "Project",
+    "Story",
+    "StoryItem",
+    "VoiceProfile",
+    # Session
+    "engine",
+    "SessionLocal",
+    "_db_path",
+    "init_db",
+    "get_db",
+]
@@ -0,0 +1,170 @@
+"""Column-level migrations for the voicebox SQLite database.
+
+Why not Alembic?  voicebox is a single-user desktop app shipping as a
+PyInstaller binary.  Every user has exactly one SQLite file.  Alembic's
+strengths -- migration tracking across environments, rollback, team
+coordination -- don't apply here and would add bundling complexity
+(alembic.ini, env.py, versions/ directory all need to survive
+PyInstaller).  The column-existence checks below are idempotent, run in
+<50 ms on startup, and have worked reliably across 12 schema changes.
+If the project ever moves to a server-based deployment or Postgres, this
+decision should be revisited.
+
+Adding a new migration:
+    1. Append a new ``_migrate_*`` helper at the bottom of this file.
+    2. Call it from ``run_migrations()`` in the appropriate spot.
+    3. The helper should check column/table existence before acting
+       (idempotent) and print a short message when it does real work.
+"""
+
+import logging
+
+from sqlalchemy import inspect, text
+
+logger = logging.getLogger(__name__)
+
+
+def run_migrations(engine) -> None:
+    """Run all schema migrations.  Safe to call on every startup."""
+    inspector = inspect(engine)
+    tables = set(inspector.get_table_names())
+
+    _migrate_story_items(engine, inspector, tables)
+    _migrate_profiles(engine, inspector, tables)
+    _migrate_generations(engine, inspector, tables)
+    _migrate_effect_presets(engine, inspector, tables)
+    _migrate_generation_versions(engine, inspector, tables)
+
+
+# -- helpers ---------------------------------------------------------------
+
+def _get_columns(inspector, table: str) -> set[str]:
+    return {col["name"] for col in inspector.get_columns(table)}
+
+
+def _add_column(engine, table: str, column_sql: str, label: str) -> None:
+    """Add a column if it doesn't already exist."""
+    with engine.connect() as conn:
+        conn.execute(text(f"ALTER TABLE {table} ADD COLUMN {column_sql}"))
+        conn.commit()
+    logger.info("Added %s column to %s", label, table)
+
+
+# -- per-table migrations --------------------------------------------------
+
+def _migrate_story_items(engine, inspector, tables: set[str]) -> None:
+    if "story_items" not in tables:
+        return
+
+    columns = _get_columns(inspector, "story_items")
+
+    # Replace position-based ordering with absolute timecodes
+    if "position" in columns:
+        logger.info("Migrating story_items: removing position column, using start_time_ms")
+        with engine.connect() as conn:
+            if "start_time_ms" not in columns:
+                conn.execute(text(
+                    "ALTER TABLE story_items ADD COLUMN start_time_ms INTEGER DEFAULT 0"
+                ))
+                result = conn.execute(text("""
+                    SELECT si.id, si.story_id, si.position, g.duration
+                    FROM story_items si
+                    JOIN generations g ON si.generation_id = g.id
+                    ORDER BY si.story_id, si.position
+                """))
+                current_story_id = None
+                current_time_ms = 0
+                for item_id, story_id, _position, duration in result.fetchall():
+                    if story_id != current_story_id:
+                        current_story_id = story_id
+                        current_time_ms = 0
+                    conn.execute(
+                        text("UPDATE story_items SET start_time_ms = :time WHERE id = :id"),
+                        {"time": current_time_ms, "id": item_id},
+                    )
+                    current_time_ms += int((duration or 0) * 1000) + 200
+                conn.commit()
+
+            # Recreate table without the position column (SQLite lacks DROP COLUMN)
+            conn.execute(text("""
+                CREATE TABLE story_items_new (
+                    id VARCHAR PRIMARY KEY,
+                    story_id VARCHAR NOT NULL,
+                    generation_id VARCHAR NOT NULL,
+                    start_time_ms INTEGER NOT NULL DEFAULT 0,
+                    track INTEGER NOT NULL DEFAULT 0,
+                    trim_start_ms INTEGER NOT NULL DEFAULT 0,
+                    trim_end_ms INTEGER NOT NULL DEFAULT 0,
+                    version_id VARCHAR,
+                    created_at DATETIME,
+                    FOREIGN KEY (story_id) REFERENCES stories(id),
+                    FOREIGN KEY (generation_id) REFERENCES generations(id)
+                )
+            """))
+            conn.execute(text("""
+                INSERT INTO story_items_new (id, story_id, generation_id, start_time_ms, track, trim_start_ms, trim_end_ms, version_id, created_at)
+                SELECT id, story_id, generation_id, start_time_ms,
+                    COALESCE(track, 0), COALESCE(trim_start_ms, 0), COALESCE(trim_end_ms, 0), version_id, created_at
+                FROM story_items
+            """))
+            conn.execute(text("DROP TABLE story_items"))
+            conn.execute(text("ALTER TABLE story_items_new RENAME TO story_items"))
+            conn.commit()
+
+        # Re-read after table recreation
+        columns = _get_columns(inspector, "story_items")
+
+    if "track" not in columns:
+        _add_column(engine, "story_items", "track INTEGER NOT NULL DEFAULT 0", "track")
+    # Re-read so subsequent checks see new columns
+    columns = _get_columns(inspector, "story_items")
+    if "trim_start_ms" not in columns:
+        _add_column(engine, "story_items", "trim_start_ms INTEGER NOT NULL DEFAULT 0", "trim_start_ms")
+    if "trim_end_ms" not in columns:
+        _add_column(engine, "story_items", "trim_end_ms INTEGER NOT NULL DEFAULT 0", "trim_end_ms")
+    if "version_id" not in columns:
+        _add_column(engine, "story_items", "version_id VARCHAR", "version_id")
+
+
+def _migrate_profiles(engine, inspector, tables: set[str]) -> None:
+    if "profiles" not in tables:
+        return
+    columns = _get_columns(inspector, "profiles")
+    if "avatar_path" not in columns:
+        _add_column(engine, "profiles", "avatar_path VARCHAR", "avatar_path")
+    if "effects_chain" not in columns:
+        _add_column(engine, "profiles", "effects_chain TEXT", "effects_chain")
+
+
+def _migrate_generations(engine, inspector, tables: set[str]) -> None:
+    if "generations" not in tables:
+        return
+    columns = _get_columns(inspector, "generations")
+    if "status" not in columns:
+        _add_column(engine, "generations", "status VARCHAR DEFAULT 'completed'", "status")
+    if "error" not in columns:
+        _add_column(engine, "generations", "error TEXT", "error")
+    if "engine" not in columns:
+        _add_column(engine, "generations", "engine VARCHAR DEFAULT 'qwen'", "engine")
+    # Re-read after engine column (variable name shadows outer scope in old code)
+    columns = _get_columns(inspector, "generations")
+    if "model_size" not in columns:
+        _add_column(engine, "generations", "model_size VARCHAR", "model_size")
+    if "is_favorited" not in columns:
+        _add_column(engine, "generations", "is_favorited BOOLEAN DEFAULT 0", "is_favorited")
+
+
+def _migrate_effect_presets(engine, inspector, tables: set[str]) -> None:
+    if "effect_presets" not in tables:
+        return
+    columns = _get_columns(inspector, "effect_presets")
+    if "sort_order" not in columns:
+        _add_column(engine, "effect_presets", "sort_order INTEGER DEFAULT 100", "sort_order")
+
+
+def _migrate_generation_versions(engine, inspector, tables: set[str]) -> None:
+    if "generation_versions" not in tables:
+        return
+    columns = _get_columns(inspector, "generation_versions")
+    if "source_version_id" not in columns:
+        _add_column(engine, "generation_versions", "source_version_id VARCHAR", "source_version_id")
@@ -0,0 +1,155 @@
+"""ORM model definitions for the voicebox SQLite database."""
+
+from datetime import datetime
+import uuid
+
+from sqlalchemy import Column, String, Integer, Float, DateTime, Text, ForeignKey, Boolean
+from sqlalchemy.ext.declarative import declarative_base
+
+Base = declarative_base()
+
+
+class VoiceProfile(Base):
+    """Voice profile."""
+
+    __tablename__ = "profiles"
+
+    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
+    name = Column(String, unique=True, nullable=False)
+    description = Column(Text)
+    language = Column(String, default="en")
+    avatar_path = Column(String, nullable=True)
+    effects_chain = Column(Text, nullable=True)
+    created_at = Column(DateTime, default=datetime.utcnow)
+    updated_at = Column(DateTime, default=datetime.utcnow, onupdate=datetime.utcnow)
+
+
+class ProfileSample(Base):
+    """Audio sample attached to a voice profile."""
+
+    __tablename__ = "profile_samples"
+
+    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
+    profile_id = Column(String, ForeignKey("profiles.id"), nullable=False)
+    audio_path = Column(String, nullable=False)
+    reference_text = Column(Text, nullable=False)
+
+
+class Generation(Base):
+    """A single TTS generation."""
+
+    __tablename__ = "generations"
+
+    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
+    profile_id = Column(String, ForeignKey("profiles.id"), nullable=False)
+    text = Column(Text, nullable=False)
+    language = Column(String, default="en")
+    audio_path = Column(String, nullable=True)
+    duration = Column(Float, nullable=True)
+    seed = Column(Integer)
+    instruct = Column(Text)
+    engine = Column(String, default="qwen")
+    model_size = Column(String, nullable=True)
+    status = Column(String, default="completed")
+    error = Column(Text, nullable=True)
+    is_favorited = Column(Boolean, default=False)
+    created_at = Column(DateTime, default=datetime.utcnow)
+
+
+class Story(Base):
+    """A story that sequences multiple generations."""
+
+    __tablename__ = "stories"
+
+    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
+    name = Column(String, nullable=False)
+    description = Column(Text)
+    created_at = Column(DateTime, default=datetime.utcnow)
+    updated_at = Column(DateTime, default=datetime.utcnow, onupdate=datetime.utcnow)
+
+
+class StoryItem(Base):
+    """Links a generation to a story at a specific timecode."""
+
+    __tablename__ = "story_items"
+
+    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
+    story_id = Column(String, ForeignKey("stories.id"), nullable=False)
+    generation_id = Column(String, ForeignKey("generations.id"), nullable=False)
+    version_id = Column(String, ForeignKey("generation_versions.id"), nullable=True)
+    start_time_ms = Column(Integer, nullable=False, default=0)
+    track = Column(Integer, nullable=False, default=0)
+    trim_start_ms = Column(Integer, nullable=False, default=0)
+    trim_end_ms = Column(Integer, nullable=False, default=0)
+    created_at = Column(DateTime, default=datetime.utcnow)
+
+
+class Project(Base):
+    """Audio studio project (JSON blob)."""
+
+    __tablename__ = "projects"
+
+    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
+    name = Column(String, nullable=False)
+    data = Column(Text)
+    created_at = Column(DateTime, default=datetime.utcnow)
+    updated_at = Column(DateTime, default=datetime.utcnow, onupdate=datetime.utcnow)
+
+
+class GenerationVersion(Base):
+    """A version of a generation's audio (original, processed, alternate takes)."""
+
+    __tablename__ = "generation_versions"
+
+    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
+    generation_id = Column(String, ForeignKey("generations.id"), nullable=False)
+    label = Column(String, nullable=False)
+    audio_path = Column(String, nullable=False)
+    effects_chain = Column(Text, nullable=True)
+    source_version_id = Column(String, ForeignKey("generation_versions.id"), nullable=True)
+    is_default = Column(Boolean, default=False)
+    created_at = Column(DateTime, default=datetime.utcnow)
+
+
+class EffectPreset(Base):
+    """Saved effect chain preset."""
+
+    __tablename__ = "effect_presets"
+
+    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
+    name = Column(String, unique=True, nullable=False)
+    description = Column(Text, nullable=True)
+    effects_chain = Column(Text, nullable=False)
+    is_builtin = Column(Boolean, default=False)
+    sort_order = Column(Integer, default=100)
+    created_at = Column(DateTime, default=datetime.utcnow)
+
+
+class AudioChannel(Base):
+    """Audio output channel (bus)."""
+
+    __tablename__ = "audio_channels"
+
+    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
+    name = Column(String, nullable=False)
+    is_default = Column(Boolean, default=False)
+    created_at = Column(DateTime, default=datetime.utcnow)
+
+
+class ChannelDeviceMapping(Base):
+    """Mapping between a channel and an OS audio device."""
+
+    __tablename__ = "channel_device_mappings"
+
+    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
+    channel_id = Column(String, ForeignKey("audio_channels.id"), nullable=False)
+    device_id = Column(String, nullable=False)
+
+
+class ProfileChannelMapping(Base):
+    """Many-to-many mapping between voice profiles and audio channels."""
+
+    __tablename__ = "profile_channel_mappings"
+
+    profile_id = Column(String, ForeignKey("profiles.id"), primary_key=True)
+    channel_id = Column(String, ForeignKey("audio_channels.id"), primary_key=True)
@@ -0,0 +1,71 @@
+"""Post-migration data seeding and backfills."""
+
+import json
+import logging
+import uuid
+from pathlib import Path
+
+logger = logging.getLogger(__name__)
+
+
+def backfill_generation_versions(SessionLocal, Generation, GenerationVersion) -> None:
+    """Create 'clean' version entries for generations that predate the versions feature."""
+    db = SessionLocal()
+    try:
+        existing_version_gen_ids = {
+            row[0] for row in db.query(GenerationVersion.generation_id).all()
+        }
+        generations = db.query(Generation).filter(
+            Generation.status == "completed",
+            Generation.audio_path.isnot(None),
+            Generation.audio_path != "",
+        ).all()
+
+        count = 0
+        for gen in generations:
+            if gen.id in existing_version_gen_ids:
+                continue
+            if not Path(gen.audio_path).exists():
+                continue
+            version = GenerationVersion(
+                id=str(uuid.uuid4()),
+                generation_id=gen.id,
+                label="clean",
+                audio_path=gen.audio_path,
+                effects_chain=None,
+                is_default=True,
+            )
+            db.add(version)
+            count += 1
+
+        if count > 0:
+            db.commit()
+            logger.info("Backfilled %d generation version entries", count)
+    finally:
+        db.close()
+
+
+def seed_builtin_presets(SessionLocal, EffectPreset) -> None:
+    """Ensure built-in effect presets exist in the database."""
+    from ..utils.effects import BUILTIN_PRESETS
+
+    db = SessionLocal()
+    try:
+        for idx, (_key, preset_data) in enumerate(BUILTIN_PRESETS.items()):
+            sort_order = preset_data.get("sort_order", idx)
+            existing = db.query(EffectPreset).filter_by(name=preset_data["name"]).first()
+            if not existing:
+                preset = EffectPreset(
+                    id=str(uuid.uuid4()),
+                    name=preset_data["name"],
+                    description=preset_data.get("description"),
+                    effects_chain=json.dumps(preset_data["effects_chain"]),
+                    is_builtin=True,
+                    sort_order=sort_order,
+                )
+                db.add(preset)
+            elif existing.sort_order != sort_order:
+                existing.sort_order = sort_order
+        db.commit()
+    finally:
+        db.close()
@@ -0,0 +1,78 @@
+"""Engine creation, initialization, and session management."""
+
+import logging
+import uuid
+
+from sqlalchemy import create_engine
+from sqlalchemy.orm import sessionmaker
+
+from .. import config
+from .models import (
+    Base,
+    AudioChannel,
+    EffectPreset,
+    Generation,
+    GenerationVersion,
+    ProfileChannelMapping,
+    VoiceProfile,
+)
+from .migrations import run_migrations
+from .seed import backfill_generation_versions, seed_builtin_presets
+
+logger = logging.getLogger(__name__)
+
+# Initialized by init_db()
+engine = None
+SessionLocal = None
+_db_path = None
+
+
+def init_db() -> None:
+    """Initialize the database engine, run migrations, create tables, and seed data."""
+    global engine, SessionLocal, _db_path
+
+    _db_path = config.get_db_path()
+    _db_path.parent.mkdir(parents=True, exist_ok=True)
+
+    engine = create_engine(
+        f"sqlite:///{_db_path}",
+        connect_args={"check_same_thread": False},
+    )
+
+    SessionLocal = sessionmaker(autocommit=False, autoflush=False, bind=engine)
+
+    run_migrations(engine)
+    Base.metadata.create_all(bind=engine)
+
+    # Create default audio channel if it doesn't exist
+    db = SessionLocal()
+    try:
+        default_channel = db.query(AudioChannel).filter(AudioChannel.is_default == True).first()
+        if not default_channel:
+            default_channel = AudioChannel(
+                id=str(uuid.uuid4()),
+                name="Default",
+                is_default=True,
+            )
+            db.add(default_channel)
+
+            for profile in db.query(VoiceProfile).all():
+                db.add(ProfileChannelMapping(
+                    profile_id=profile.id,
+                    channel_id=default_channel.id,
+                ))
+            db.commit()
+    finally:
+        db.close()
+
+    backfill_generation_versions(SessionLocal, Generation, GenerationVersion)
+    seed_builtin_presets(SessionLocal, EffectPreset)
+
+
+def get_db():
+    """Yield a database session (FastAPI dependency)."""
+    db = SessionLocal()
+    try:
+        yield db
+    finally:
+        db.close()
@@ -1,221 +0,0 @@
-"""
-Example usage of the voicebox backend API.
-
-This script demonstrates how to:
-1. Create a voice profile
-2. Add samples to the profile
-3. Generate speech
-4. List history
-"""
-
-import requests
-import time
-from pathlib import Path
-
-# API base URL
-BASE_URL = "http://localhost:8000"
-
-
-def check_health():
-    """Check if the server is running."""
-    response = requests.get(f"{BASE_URL}/health")
-    data = response.json()
-    print(f"Server status: {data['status']}")
-    print(f"Model loaded: {data['model_loaded']}")
-    print(f"GPU available: {data['gpu_available']}")
-    print()
-    return data
-
-
-def create_profile(name: str, description: str = None, language: str = "en"):
-    """Create a new voice profile."""
-    response = requests.post(
-        f"{BASE_URL}/profiles",
-        json={
-            "name": name,
-            "description": description,
-            "language": language,
-        },
-    )
-    response.raise_for_status()
-    profile = response.json()
-    print(f"Created profile: {profile['name']} (ID: {profile['id']})")
-    return profile
-
-
-def add_sample(profile_id: str, audio_file: str, reference_text: str):
-    """Add a sample to a voice profile."""
-    with open(audio_file, "rb") as f:
-        files = {"file": f}
-        data = {"reference_text": reference_text}
-        response = requests.post(
-            f"{BASE_URL}/profiles/{profile_id}/samples",
-            files=files,
-            data=data,
-        )
-    response.raise_for_status()
-    sample = response.json()
-    print(f"Added sample: {sample['id']}")
-    return sample
-
-
-def generate_speech(profile_id: str, text: str, language: str = "en", seed: int = None):
-    """Generate speech using a voice profile."""
-    print(f"Generating speech: '{text[:50]}...'")
-    start_time = time.time()
-    
-    response = requests.post(
-        f"{BASE_URL}/generate",
-        json={
-            "profile_id": profile_id,
-            "text": text,
-            "language": language,
-            "seed": seed,
-        },
-    )
-    response.raise_for_status()
-    generation = response.json()
-    
-    elapsed = time.time() - start_time
-    print(f"Generated in {elapsed:.2f}s (duration: {generation['duration']:.2f}s)")
-    print(f"Generation ID: {generation['id']}")
-    return generation
-
-
-def download_audio(generation_id: str, output_file: str):
-    """Download generated audio."""
-    response = requests.get(f"{BASE_URL}/audio/{generation_id}")
-    response.raise_for_status()
-    
-    with open(output_file, "wb") as f:
-        f.write(response.content)
-    
-    print(f"Saved audio to: {output_file}")
-
-
-def list_profiles():
-    """List all voice profiles."""
-    response = requests.get(f"{BASE_URL}/profiles")
-    response.raise_for_status()
-    profiles = response.json()
-    
-    print(f"Found {len(profiles)} profiles:")
-    for profile in profiles:
-        print(f"  - {profile['name']} (ID: {profile['id']})")
-    
-    return profiles
-
-
-def list_history(profile_id: str = None, limit: int = 10):
-    """List generation history."""
-    params = {"limit": limit}
-    if profile_id:
-        params["profile_id"] = profile_id
-    
-    response = requests.get(f"{BASE_URL}/history", params=params)
-    response.raise_for_status()
-    history = response.json()
-    
-    print(f"Found {len(history)} generations:")
-    for gen in history:
-        print(f"  - {gen['text'][:50]}... ({gen['duration']:.2f}s)")
-    
-    return history
-
-
-def transcribe_audio(audio_file: str, language: str = None):
-    """Transcribe audio file."""
-    print(f"Transcribing: {audio_file}")
-    
-    with open(audio_file, "rb") as f:
-        files = {"file": f}
-        data = {}
-        if language:
-            data["language"] = language
-        
-        response = requests.post(
-            f"{BASE_URL}/transcribe",
-            files=files,
-            data=data,
-        )
-    
-    response.raise_for_status()
-    result = response.json()
-    
-    print(f"Transcription: {result['text']}")
-    print(f"Duration: {result['duration']:.2f}s")
-    return result
-
-
-def main():
-    """Run example workflow."""
-    print("=" * 60)
-    print("voicebox Backend API Example")
-    print("=" * 60)
-    print()
-    
-    # 1. Check health
-    print("1. Checking server health...")
-    check_health()
-    
-    # 2. Create a profile
-    print("2. Creating voice profile...")
-    profile = create_profile(
-        name="Example Voice",
-        description="A test voice profile",
-        language="en",
-    )
-    profile_id = profile["id"]
-    print()
-    
-    # 3. Add samples (you'll need actual audio files)
-    print("3. Adding samples...")
-    print("   (Skipping - add your own audio files here)")
-    # Uncomment and add your audio file:
-    # sample = add_sample(
-    #     profile_id,
-    #     "path/to/your/sample.wav",
-    #     "This is the transcript of the audio",
-    # )
-    print()
-    
-    # 4. Generate speech (requires samples to be added first)
-    print("4. Generating speech...")
-    print("   (Skipping - add samples first)")
-    # Uncomment after adding samples:
-    # generation = generate_speech(
-    #     profile_id,
-    #     "Hello, this is a test of the voice cloning system.",
-    #     language="en",
-    #     seed=42,
-    # )
-    # 
-    # # 5. Download audio
-    # print("\n5. Downloading audio...")
-    # download_audio(generation["id"], "output.wav")
-    print()
-    
-    # 6. List profiles
-    print("6. Listing all profiles...")
-    list_profiles()
-    print()
-    
-    # 7. List history
-    print("7. Listing generation history...")
-    list_history(limit=5)
-    print()
-    
-    # 8. Transcribe audio (you'll need an audio file)
-    print("8. Transcribing audio...")
-    print("   (Skipping - add your own audio file here)")
-    # Uncomment and add your audio file:
-    # transcribe_audio("path/to/audio.wav", language="en")
-    print()
-    
-    print("=" * 60)
-    print("Example complete!")
-    print("=" * 60)
-
-
-if __name__ == "__main__":
-    main()
@@ -1,48 +0,0 @@
-"""
-Database migration script to add instruct column to generations table.
-
-Run this once to update existing databases:
-    python -m backend.migrate_add_instruct
-"""
-
-import sqlite3
-import os
-from pathlib import Path
-
-
-def migrate():
-    """Add instruct column to generations table if it doesn't exist."""
-    # Get data directory
-    data_dir = os.environ.get("VOICEBOX_DATA_DIR")
-    if data_dir:
-        db_path = Path(data_dir) / "voicebox.db"
-    else:
-        db_path = Path.cwd() / "data" / "voicebox.db"
-
-    if not db_path.exists():
-        print(f"Database not found at {db_path}, skipping migration")
-        return
-
-    conn = sqlite3.connect(db_path)
-    cursor = conn.cursor()
-
-    # Check if instruct column already exists
-    cursor.execute("PRAGMA table_info(generations)")
-    columns = [row[1] for row in cursor.fetchall()]
-
-    if 'instruct' in columns:
-        print("instruct column already exists, skipping migration")
-        conn.close()
-        return
-
-    # Add instruct column
-    print("Adding instruct column to generations table...")
-    cursor.execute("ALTER TABLE generations ADD COLUMN instruct TEXT")
-    conn.commit()
-    conn.close()
-
-    print("Migration complete!")
-
-
-if __name__ == "__main__":
-    migrate()
@@ -9,13 +9,17 @@ from datetime import datetime

 class VoiceProfileCreate(BaseModel):
    """Request model for creating a voice profile."""
+
    name: str = Field(..., min_length=1, max_length=100)
    description: Optional[str] = Field(None, max_length=500)
-    language: str = Field(default="en", pattern="^(zh|en|ja|ko|de|fr|ru|pt|es|it|he|ar|da|el|fi|hi|ms|nl|no|pl|sv|sw|tr)$")
+    language: str = Field(
+        default="en", pattern="^(zh|en|ja|ko|de|fr|ru|pt|es|it|he|ar|da|el|fi|hi|ms|nl|no|pl|sv|sw|tr)$"
+    )


 class VoiceProfileResponse(BaseModel):
    """Response model for voice profile."""
+
    id: str
    name: str
    description: Optional[str]
@@ -33,16 +37,19 @@ class VoiceProfileResponse(BaseModel):

 class ProfileSampleCreate(BaseModel):
    """Request model for adding a sample to a profile."""
+
    reference_text: str = Field(..., min_length=1, max_length=1000)


 class ProfileSampleUpdate(BaseModel):
    """Request model for updating a profile sample."""
+
    reference_text: str = Field(..., min_length=1, max_length=1000)


 class ProfileSampleResponse(BaseModel):
    """Response model for profile sample."""
+
    id: str
    profile_id: str
    audio_path: str
@@ -54,21 +61,29 @@ class ProfileSampleResponse(BaseModel):

 class GenerationRequest(BaseModel):
    """Request model for voice generation."""
+
    profile_id: str
    text: str = Field(..., min_length=1, max_length=50000)
-    language: str = Field(default="en", pattern="^(zh|en|ja|ko|de|fr|ru|pt|es|it|he)$")
+    language: str = Field(default="en", pattern="^(zh|en|ja|ko|de|fr|ru|pt|es|it|he|ar|da|el|fi|hi|ms|nl|no|pl|sv|sw|tr)$")
    seed: Optional[int] = Field(None, ge=0)
    model_size: Optional[str] = Field(default="1.7B", pattern="^(1\\.7B|0\\.6B)$")
    instruct: Optional[str] = Field(None, max_length=500)
    engine: Optional[str] = Field(default="qwen", pattern="^(qwen|luxtts|chatterbox|chatterbox_turbo)$")
-    max_chunk_chars: int = Field(default=800, ge=100, le=5000, description="Max characters per chunk for long text splitting")
-    crossfade_ms: int = Field(default=50, ge=0, le=500, description="Crossfade duration in ms between chunks (0 for hard cut)")
+    max_chunk_chars: int = Field(
+        default=800, ge=100, le=5000, description="Max characters per chunk for long text splitting"
+    )
+    crossfade_ms: int = Field(
+        default=50, ge=0, le=500, description="Crossfade duration in ms between chunks (0 for hard cut)"
+    )
    normalize: bool = Field(default=True, description="Normalize output audio volume")
-    effects_chain: Optional[List["EffectConfig"]] = Field(None, description="Effects chain to apply after generation (overrides profile default)")
+    effects_chain: Optional[List["EffectConfig"]] = Field(
+        None, description="Effects chain to apply after generation (overrides profile default)"
+    )


 class GenerationResponse(BaseModel):
    """Response model for voice generation."""
+
    id: str
    profile_id: str
    text: str
@@ -92,6 +107,7 @@ class GenerationResponse(BaseModel):

 class HistoryQuery(BaseModel):
    """Query model for generation history."""
+
    profile_id: Optional[str] = None
    search: Optional[str] = None
    limit: int = Field(default=50, ge=1, le=100)
@@ -100,6 +116,7 @@ class HistoryQuery(BaseModel):

 class HistoryResponse(BaseModel):
    """Response model for history entry (includes profile name)."""
+
    id: str
    profile_id: str
    profile_name: str
@@ -124,23 +141,27 @@ class HistoryResponse(BaseModel):

 class HistoryListResponse(BaseModel):
    """Response model for history list."""
+
    items: List[HistoryResponse]
    total: int


 class TranscriptionRequest(BaseModel):
    """Request model for audio transcription."""
+
    language: Optional[str] = Field(None, pattern="^(en|zh)$")


 class TranscriptionResponse(BaseModel):
    """Response model for transcription."""
+
    text: str
    duration: float


 class HealthResponse(BaseModel):
    """Response model for health check."""
+
    status: str
    model_loaded: bool
    model_downloaded: Optional[bool] = None  # Whether model is cached/downloaded
@@ -154,6 +175,7 @@ class HealthResponse(BaseModel):

 class DirectoryCheck(BaseModel):
    """Health status for a single directory."""
+
    path: str
    exists: bool
    writable: bool
@@ -162,6 +184,7 @@ class DirectoryCheck(BaseModel):

 class FilesystemHealthResponse(BaseModel):
    """Response model for filesystem health check."""
+
    healthy: bool
    disk_free_mb: Optional[float] = None
    disk_total_mb: Optional[float] = None
@@ -170,6 +193,7 @@ class FilesystemHealthResponse(BaseModel):

 class ModelStatus(BaseModel):
    """Response model for model status."""
+
    model_name: str
    display_name: str
    hf_repo_id: Optional[str] = None  # HuggingFace repository ID
@@ -181,33 +205,38 @@ class ModelStatus(BaseModel):

 class ModelStatusListResponse(BaseModel):
    """Response model for model status list."""
+
    models: List[ModelStatus]


 class ModelDownloadRequest(BaseModel):
    """Request model for triggering model download."""
+
    model_name: str


 class ModelMigrateRequest(BaseModel):
    """Request model for migrating models to a new directory."""
+
    destination: str


 class ActiveDownloadTask(BaseModel):
    """Response model for active download task."""
+
    model_name: str
    status: str
    started_at: datetime
    error: Optional[str] = None
    progress: Optional[float] = None  # 0-100 percentage
-    current: Optional[int] = None     # bytes downloaded
-    total: Optional[int] = None       # total bytes
-    filename: Optional[str] = None    # current file being downloaded
+    current: Optional[int] = None  # bytes downloaded
+    total: Optional[int] = None  # total bytes
+    filename: Optional[str] = None  # current file being downloaded


 class ActiveGenerationTask(BaseModel):
    """Response model for active generation task."""
+
    task_id: str
    profile_id: str
    text_preview: str
@@ -216,24 +245,28 @@ class ActiveGenerationTask(BaseModel):

 class ActiveTasksResponse(BaseModel):
    """Response model for active tasks."""
+
    downloads: List[ActiveDownloadTask]
    generations: List[ActiveGenerationTask]


 class AudioChannelCreate(BaseModel):
    """Request model for creating an audio channel."""
+
    name: str = Field(..., min_length=1, max_length=100)
    device_ids: List[str] = Field(default_factory=list)


 class AudioChannelUpdate(BaseModel):
    """Request model for updating an audio channel."""
+
    name: Optional[str] = Field(None, min_length=1, max_length=100)
    device_ids: Optional[List[str]] = None


 class AudioChannelResponse(BaseModel):
    """Response model for audio channel."""
+
    id: str
    name: str
    is_default: bool
@@ -246,22 +279,26 @@ class AudioChannelResponse(BaseModel):

 class ChannelVoiceAssignment(BaseModel):
    """Request model for assigning voices to a channel."""
+
    profile_ids: List[str]


 class ProfileChannelAssignment(BaseModel):
    """Request model for assigning channels to a profile."""
+
    channel_ids: List[str]


 class StoryCreate(BaseModel):
    """Request model for creating a story."""
+
    name: str = Field(..., min_length=1, max_length=100)
    description: Optional[str] = Field(None, max_length=500)


 class StoryResponse(BaseModel):
    """Response model for story (list view)."""
+
    id: str
    name: str
    description: Optional[str]
@@ -275,6 +312,7 @@ class StoryResponse(BaseModel):

 class StoryItemDetail(BaseModel):
    """Detail model for story item with generation info."""
+
    id: str
    story_id: str
    generation_id: str
@@ -304,6 +342,7 @@ class StoryItemDetail(BaseModel):

 class StoryDetailResponse(BaseModel):
    """Response model for story with items."""
+
    id: str
    name: str
    description: Optional[str]
@@ -317,6 +356,7 @@ class StoryDetailResponse(BaseModel):

 class StoryItemCreate(BaseModel):
    """Request model for adding a generation to a story."""
+
    generation_id: str
    start_time_ms: Optional[int] = None  # If not provided, will be calculated automatically
    track: Optional[int] = 0  # Track number (0 = main track)
@@ -324,48 +364,52 @@ class StoryItemCreate(BaseModel):

 class StoryItemUpdateTime(BaseModel):
    """Request model for updating a story item's timecode."""
+
    generation_id: str
    start_time_ms: int = Field(..., ge=0)


 class StoryItemBatchUpdate(BaseModel):
    """Request model for batch updating story item timecodes."""
+
    updates: List[StoryItemUpdateTime]


 class StoryItemReorder(BaseModel):
    """Request model for reordering story items."""
+
    generation_ids: List[str] = Field(..., min_length=1)


 class StoryItemMove(BaseModel):
    """Request model for moving a story item (position and/or track)."""
+
    start_time_ms: int = Field(..., ge=0)
    track: int = 0


 class StoryItemTrim(BaseModel):
    """Request model for trimming a story item."""
+
    trim_start_ms: int = Field(..., ge=0)
    trim_end_ms: int = Field(..., ge=0)


 class StoryItemSplit(BaseModel):
    """Request model for splitting a story item."""
+
    split_time_ms: int = Field(..., ge=0)  # Time within the clip to split at (relative to clip start)


 class StoryItemVersionUpdate(BaseModel):
    """Request model for setting a story item's pinned version."""
+
    version_id: Optional[str] = None  # null = use generation default


-# ============================================
-# Effects & Versions
-# ============================================
-
 class EffectConfig(BaseModel):
    """A single effect in an effects chain."""
+
    type: str
    enabled: bool = True
    params: dict = Field(default_factory=dict)
@@ -373,11 +417,13 @@ class EffectConfig(BaseModel):

 class EffectsChain(BaseModel):
    """An ordered list of effects to apply."""
+
    effects: List[EffectConfig] = Field(default_factory=list)


 class EffectPresetCreate(BaseModel):
    """Request model for creating an effect preset."""
+
    name: str = Field(..., min_length=1, max_length=100)
    description: Optional[str] = Field(None, max_length=500)
    effects_chain: List[EffectConfig]
@@ -385,6 +431,7 @@ class EffectPresetCreate(BaseModel):

 class EffectPresetUpdate(BaseModel):
    """Request model for updating an effect preset."""
+
    name: Optional[str] = Field(None, min_length=1, max_length=100)
    description: Optional[str] = None
    effects_chain: Optional[List[EffectConfig]] = None
@@ -392,6 +439,7 @@ class EffectPresetUpdate(BaseModel):

 class EffectPresetResponse(BaseModel):
    """Response model for effect preset."""
+
    id: str
    name: str
    description: Optional[str] = None
@@ -405,6 +453,7 @@ class EffectPresetResponse(BaseModel):

 class GenerationVersionResponse(BaseModel):
    """Response model for a generation version."""
+
    id: str
    generation_id: str
    label: str
@@ -420,19 +469,24 @@ class GenerationVersionResponse(BaseModel):

 class ApplyEffectsRequest(BaseModel):
    """Request to apply effects to an existing generation."""
+
    effects_chain: List[EffectConfig]
-    source_version_id: Optional[str] = Field(None, description="Version to use as source audio (defaults to clean/original)")
+    source_version_id: Optional[str] = Field(
+        None, description="Version to use as source audio (defaults to clean/original)"
+    )
    label: Optional[str] = Field(None, max_length=100, description="Label for this version (auto-generated if omitted)")
    set_as_default: bool = Field(default=True, description="Set this version as the default")


 class ProfileEffectsUpdate(BaseModel):
    """Request to update the default effects chain on a profile."""
+
    effects_chain: Optional[List[EffectConfig]] = Field(None, description="Effects chain (null to remove)")


 class AvailableEffectParam(BaseModel):
    """Description of a single effect parameter."""
+
    default: float
    min: float
    max: float
@@ -442,6 +496,7 @@ class AvailableEffectParam(BaseModel):

 class AvailableEffect(BaseModel):
    """Description of an available effect type."""
+
    type: str
    label: str
    description: str
@@ -450,4 +505,5 @@ class AvailableEffect(BaseModel):

 class AvailableEffectsResponse(BaseModel):
    """Response listing all available effect types."""
+
    effects: List[AvailableEffect]
@@ -0,0 +1,83 @@
+[project]
+name = "voicebox-backend"
+version = "0.2.3"
+requires-python = ">=3.12"
+
+# ---------------------------------------------------------------------------
+# Ruff – linter + formatter
+# ---------------------------------------------------------------------------
+
+[tool.ruff]
+target-version = "py312"
+line-length = 120
+src = ["."]
+
+# Files/dirs to skip entirely.
+extend-exclude = [
+    "voicebox-server.spec",
+    "build_binary.py",
+]
+
+[tool.ruff.lint]
+select = [
+    "F",     # pyflakes
+    "E",     # pycodestyle errors
+    "W",     # pycodestyle warnings
+    "I",     # isort
+    "N",     # pep8-naming
+    "UP",    # pyupgrade (modernize syntax for 3.12)
+    "B",     # flake8-bugbear
+    "A",     # flake8-builtins (shadowing built-in names)
+    "SIM",   # flake8-simplify
+    "T20",   # flake8-print (flag print() calls)
+    "RET",   # flake8-return
+    "PIE",   # misc lints
+    "PT",    # flake8-pytest-style
+    "RUF",   # ruff-specific rules
+    "ERA",   # commented-out code detection
+    "FIX",   # flag TODO/FIXME/HACK/XXX for review
+]
+
+ignore = [
+    # Allow print() in existing code -- remove items from this list as files
+    # are migrated to logging during the refactor.
+    "T201",   # print() found
+
+    # These conflict with the formatter or are too noisy during migration:
+    "E501",   # line too long (formatter handles this)
+    "RET504", # unnecessary assignment before return
+    "SIM108", # use ternary operator (sometimes less readable)
+    "B008",   # function call in default argument (FastAPI Depends() pattern)
+    "UP007",  # use X | Y for union (auto-fixed by UP, but noisy on big diffs)
+]
+
+# Per-file rule overrides.
+[tool.ruff.lint.per-file-ignores]
+# Tests can use assert, print, and magic values freely.
+"tests/**" = ["S101", "T201", "PLR2004", "ERA001"]
+# __init__.py re-exports are expected to have unused imports.
+"**/__init__.py" = ["F401"]
+# Entry points and scripts legitimately use print.
+"server.py" = ["T201"]
+"main.py" = ["T201"]
+# AMD GPU env vars must be set before torch import.
+"app.py" = ["E402"]
+
+[tool.ruff.lint.isort]
+known-first-party = ["backend"]
+# Group "from backend.*" imports into the first-party section.
+force-single-line = false
+combine-as-imports = true
+
+[tool.ruff.format]
+quote-style = "double"
+indent-style = "space"
+docstring-code-format = true
+
+# ---------------------------------------------------------------------------
+# pytest
+# ---------------------------------------------------------------------------
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+asyncio_mode = "auto"
@@ -0,0 +1,32 @@
+"""Route registration for the voicebox API."""
+
+from fastapi import FastAPI
+
+
+def register_routers(app: FastAPI) -> None:
+    """Include all domain routers on the application."""
+    from .health import router as health_router
+    from .profiles import router as profiles_router
+    from .channels import router as channels_router
+    from .generations import router as generations_router
+    from .history import router as history_router
+    from .transcription import router as transcription_router
+    from .stories import router as stories_router
+    from .effects import router as effects_router
+    from .audio import router as audio_router
+    from .models import router as models_router
+    from .tasks import router as tasks_router
+    from .cuda import router as cuda_router
+
+    app.include_router(health_router)
+    app.include_router(profiles_router)
+    app.include_router(channels_router)
+    app.include_router(generations_router)
+    app.include_router(history_router)
+    app.include_router(transcription_router)
+    app.include_router(stories_router)
+    app.include_router(effects_router)
+    app.include_router(audio_router)
+    app.include_router(models_router)
+    app.include_router(tasks_router)
+    app.include_router(cuda_router)
@@ -0,0 +1,71 @@
+"""Audio file serving endpoints."""
+
+from pathlib import Path
+
+from fastapi import APIRouter, Depends, HTTPException
+from fastapi.responses import FileResponse
+from sqlalchemy.orm import Session
+
+from .. import models
+from ..services import history
+from ..database import get_db
+
+router = APIRouter()
+
+
+@router.get("/audio/version/{version_id}")
+async def get_version_audio(version_id: str, db: Session = Depends(get_db)):
+    """Serve audio for a specific version."""
+    from ..services import versions as versions_mod
+
+    version = versions_mod.get_version(version_id, db)
+    if not version:
+        raise HTTPException(status_code=404, detail="Version not found")
+
+    audio_path = Path(version.audio_path)
+    if not audio_path.exists():
+        raise HTTPException(status_code=404, detail="Audio file not found")
+
+    return FileResponse(
+        audio_path,
+        media_type="audio/wav",
+        filename=f"generation_{version.generation_id}_{version.label}.wav",
+    )
+
+
+@router.get("/audio/{generation_id}")
+async def get_audio(generation_id: str, db: Session = Depends(get_db)):
+    """Serve generated audio file (serves the default version)."""
+    generation = await history.get_generation(generation_id, db)
+    if not generation:
+        raise HTTPException(status_code=404, detail="Generation not found")
+
+    audio_path = Path(generation.audio_path)
+    if not audio_path.exists():
+        raise HTTPException(status_code=404, detail="Audio file not found")
+
+    return FileResponse(
+        audio_path,
+        media_type="audio/wav",
+        filename=f"generation_{generation_id}.wav",
+    )
+
+
+@router.get("/samples/{sample_id}")
+async def get_sample_audio(sample_id: str, db: Session = Depends(get_db)):
+    """Serve profile sample audio file."""
+    from ..database import ProfileSample as DBProfileSample
+
+    sample = db.query(DBProfileSample).filter_by(id=sample_id).first()
+    if not sample:
+        raise HTTPException(status_code=404, detail="Sample not found")
+
+    audio_path = Path(sample.audio_path)
+    if not audio_path.exists():
+        raise HTTPException(status_code=404, detail="Audio file not found")
+
+    return FileResponse(
+        audio_path,
+        media_type="audio/wav",
+        filename=f"sample_{sample_id}.wav",
+    )
@@ -0,0 +1,98 @@
+"""Audio channel endpoints."""
+
+from fastapi import APIRouter, Depends, HTTPException
+from sqlalchemy.orm import Session
+
+from .. import models
+from ..services import channels
+from ..database import get_db
+
+router = APIRouter()
+
+
+@router.get("/channels", response_model=list[models.AudioChannelResponse])
+async def list_channels(db: Session = Depends(get_db)):
+    """List all audio channels."""
+    return await channels.list_channels(db)
+
+
+@router.post("/channels", response_model=models.AudioChannelResponse)
+async def create_channel(
+    data: models.AudioChannelCreate,
+    db: Session = Depends(get_db),
+):
+    """Create a new audio channel."""
+    try:
+        return await channels.create_channel(data, db)
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+
+
+@router.get("/channels/{channel_id}", response_model=models.AudioChannelResponse)
+async def get_channel(
+    channel_id: str,
+    db: Session = Depends(get_db),
+):
+    """Get an audio channel by ID."""
+    channel = await channels.get_channel(channel_id, db)
+    if not channel:
+        raise HTTPException(status_code=404, detail="Channel not found")
+    return channel
+
+
+@router.put("/channels/{channel_id}", response_model=models.AudioChannelResponse)
+async def update_channel(
+    channel_id: str,
+    data: models.AudioChannelUpdate,
+    db: Session = Depends(get_db),
+):
+    """Update an audio channel."""
+    try:
+        channel = await channels.update_channel(channel_id, data, db)
+        if not channel:
+            raise HTTPException(status_code=404, detail="Channel not found")
+        return channel
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+
+
+@router.delete("/channels/{channel_id}")
+async def delete_channel(
+    channel_id: str,
+    db: Session = Depends(get_db),
+):
+    """Delete an audio channel."""
+    try:
+        success = await channels.delete_channel(channel_id, db)
+        if not success:
+            raise HTTPException(status_code=404, detail="Channel not found")
+        return {"message": "Channel deleted successfully"}
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+
+
+@router.get("/channels/{channel_id}/voices")
+async def get_channel_voices(
+    channel_id: str,
+    db: Session = Depends(get_db),
+):
+    """Get list of profile IDs assigned to a channel."""
+    try:
+        profile_ids = await channels.get_channel_voices(channel_id, db)
+        return {"profile_ids": profile_ids}
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+
+
+@router.put("/channels/{channel_id}/voices")
+async def set_channel_voices(
+    channel_id: str,
+    data: models.ChannelVoiceAssignment,
+    db: Session = Depends(get_db),
+):
+    """Set which voices are assigned to a channel."""
+    try:
+        await channels.set_channel_voices(channel_id, data, db)
+        return {"message": "Channel voices updated successfully"}
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
@@ -0,0 +1,82 @@
+"""CUDA backend management endpoints."""
+
+import logging
+
+from fastapi import APIRouter, HTTPException
+from fastapi.responses import StreamingResponse
+
+from ..services.task_queue import create_background_task
+from ..utils.progress import get_progress_manager
+
+router = APIRouter()
+
+logger = logging.getLogger(__name__)
+
+
+@router.get("/backend/cuda-status")
+async def get_cuda_status():
+    """Get CUDA backend download/availability status."""
+    from ..services import cuda
+
+    return cuda.get_cuda_status()
+
+
+@router.post("/backend/download-cuda")
+async def download_cuda_backend():
+    """Download the CUDA backend binary."""
+    from ..services import cuda
+
+    if cuda.get_cuda_binary_path() is not None:
+        raise HTTPException(status_code=409, detail="CUDA backend already downloaded")
+
+    progress_manager = get_progress_manager()
+    existing = progress_manager.get_progress(cuda.PROGRESS_KEY)
+    if existing and existing.get("status") == "downloading":
+        raise HTTPException(status_code=409, detail="CUDA backend download already in progress")
+
+    async def _download():
+        try:
+            await cuda.download_cuda_binary()
+        except Exception as e:
+            logger.error("CUDA download failed: %s", e)
+
+    create_background_task(_download())
+    return {"message": "CUDA backend download started", "progress_key": "cuda-backend"}
+
+
+@router.delete("/backend/cuda")
+async def delete_cuda_backend():
+    """Delete the downloaded CUDA backend binary."""
+    from ..services import cuda
+
+    if cuda.is_cuda_active():
+        raise HTTPException(
+            status_code=409,
+            detail="Cannot delete CUDA backend while it is active. Switch to CPU first.",
+        )
+
+    deleted = await cuda.delete_cuda_binary()
+    if not deleted:
+        raise HTTPException(status_code=404, detail="No CUDA backend found to delete")
+
+    return {"message": "CUDA backend deleted"}
+
+
+@router.get("/backend/cuda-progress")
+async def get_cuda_download_progress():
+    """Get CUDA backend download progress via Server-Sent Events."""
+    progress_manager = get_progress_manager()
+
+    async def event_generator():
+        async for event in progress_manager.subscribe("cuda-backend"):
+            yield event
+
+    return StreamingResponse(
+        event_generator(),
+        media_type="text/event-stream",
+        headers={
+            "Cache-Control": "no-cache",
+            "Connection": "keep-alive",
+            "X-Accel-Buffering": "no",
+        },
+    )
@@ -0,0 +1,261 @@
+"""Effects presets and generation version endpoints."""
+
+import asyncio
+import io
+import uuid
+from pathlib import Path
+
+from fastapi import APIRouter, Depends, HTTPException
+from fastapi.responses import StreamingResponse
+from sqlalchemy.orm import Session
+
+from .. import config, models
+from ..services import history
+from ..database import Generation as DBGeneration, get_db
+
+router = APIRouter()
+
+
+@router.post("/effects/preview/{generation_id}")
+async def preview_effects(
+    generation_id: str,
+    data: models.ApplyEffectsRequest,
+    db: Session = Depends(get_db),
+):
+    """Apply effects to a generation's clean audio and stream back without saving."""
+    gen = db.query(DBGeneration).filter_by(id=generation_id).first()
+    if not gen:
+        raise HTTPException(status_code=404, detail="Generation not found")
+    if (gen.status or "completed") != "completed":
+        raise HTTPException(status_code=400, detail="Generation is not completed")
+
+    from ..services import versions as versions_mod
+    from ..utils.effects import apply_effects, validate_effects_chain
+    from ..utils.audio import load_audio
+
+    chain_dicts = [e.model_dump() for e in data.effects_chain]
+    error = validate_effects_chain(chain_dicts)
+    if error:
+        raise HTTPException(status_code=400, detail=error)
+
+    all_versions = versions_mod.list_versions(generation_id, db)
+    clean_version = next((v for v in all_versions if v.effects_chain is None), None)
+    source_path = clean_version.audio_path if clean_version else gen.audio_path
+    if not source_path or not Path(source_path).exists():
+        raise HTTPException(status_code=404, detail="Source audio file not found")
+
+    audio, sample_rate = await asyncio.to_thread(load_audio, source_path)
+    processed = await asyncio.to_thread(apply_effects, audio, sample_rate, chain_dicts)
+
+    import soundfile as sf
+
+    buf = io.BytesIO()
+    await asyncio.to_thread(lambda: sf.write(buf, processed, sample_rate, format="WAV"))
+    buf.seek(0)
+
+    return StreamingResponse(
+        buf,
+        media_type="audio/wav",
+        headers={
+            "Content-Disposition": f'inline; filename="preview_{generation_id}.wav"',
+            "Cache-Control": "no-cache, no-store",
+        },
+    )
+
+
+@router.get("/effects/available", response_model=models.AvailableEffectsResponse)
+async def get_available_effects():
+    """List all available effect types with parameter definitions."""
+    from ..utils.effects import get_available_effects as _get_effects
+
+    return models.AvailableEffectsResponse(effects=[models.AvailableEffect(**e) for e in _get_effects()])
+
+
+@router.get("/effects/presets", response_model=list[models.EffectPresetResponse])
+async def list_effect_presets(db: Session = Depends(get_db)):
+    """List all effect presets (built-in + user-created)."""
+    from ..services import effects as effects_mod
+
+    return effects_mod.list_presets(db)
+
+
+@router.get("/effects/presets/{preset_id}", response_model=models.EffectPresetResponse)
+async def get_effect_preset(preset_id: str, db: Session = Depends(get_db)):
+    """Get a specific effect preset."""
+    from ..services import effects as effects_mod
+
+    preset = effects_mod.get_preset(preset_id, db)
+    if not preset:
+        raise HTTPException(status_code=404, detail="Preset not found")
+    return preset
+
+
+@router.post("/effects/presets", response_model=models.EffectPresetResponse)
+async def create_effect_preset(
+    data: models.EffectPresetCreate,
+    db: Session = Depends(get_db),
+):
+    """Create a new effect preset."""
+    from ..services import effects as effects_mod
+
+    try:
+        return effects_mod.create_preset(data, db)
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+
+
+@router.put("/effects/presets/{preset_id}", response_model=models.EffectPresetResponse)
+async def update_effect_preset(
+    preset_id: str,
+    data: models.EffectPresetUpdate,
+    db: Session = Depends(get_db),
+):
+    """Update an effect preset."""
+    from ..services import effects as effects_mod
+
+    try:
+        result = effects_mod.update_preset(preset_id, data, db)
+        if not result:
+            raise HTTPException(status_code=404, detail="Preset not found")
+        return result
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+
+
+@router.delete("/effects/presets/{preset_id}")
+async def delete_effect_preset(preset_id: str, db: Session = Depends(get_db)):
+    """Delete a user effect preset."""
+    from ..services import effects as effects_mod
+
+    try:
+        if not effects_mod.delete_preset(preset_id, db):
+            raise HTTPException(status_code=404, detail="Preset not found")
+        return {"status": "deleted"}
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+
+
+@router.get(
+    "/generations/{generation_id}/versions",
+    response_model=list[models.GenerationVersionResponse],
+)
+async def list_generation_versions(
+    generation_id: str,
+    db: Session = Depends(get_db),
+):
+    """List all versions for a generation."""
+    gen = await history.get_generation(generation_id, db)
+    if not gen:
+        raise HTTPException(status_code=404, detail="Generation not found")
+
+    from ..services import versions as versions_mod
+
+    return versions_mod.list_versions(generation_id, db)
+
+
+@router.post(
+    "/generations/{generation_id}/versions/apply-effects",
+    response_model=models.GenerationVersionResponse,
+)
+async def apply_effects_to_generation(
+    generation_id: str,
+    data: models.ApplyEffectsRequest,
+    db: Session = Depends(get_db),
+):
+    """Apply an effects chain to an existing generation, creating a new version."""
+    gen = db.query(DBGeneration).filter_by(id=generation_id).first()
+    if not gen:
+        raise HTTPException(status_code=404, detail="Generation not found")
+    if (gen.status or "completed") != "completed":
+        raise HTTPException(status_code=400, detail="Generation is not completed")
+
+    from ..services import versions as versions_mod
+    from ..utils.effects import apply_effects, validate_effects_chain
+    from ..utils.audio import load_audio, save_audio
+
+    chain_dicts = [e.model_dump() for e in data.effects_chain]
+    error = validate_effects_chain(chain_dicts)
+    if error:
+        raise HTTPException(status_code=400, detail=error)
+
+    all_versions = versions_mod.list_versions(generation_id, db)
+    source_version_id = data.source_version_id
+    if source_version_id:
+        source_version = next((v for v in all_versions if v.id == source_version_id), None)
+        if not source_version:
+            raise HTTPException(status_code=404, detail="Source version not found")
+        source_path = source_version.audio_path
+    else:
+        clean_version = next((v for v in all_versions if v.effects_chain is None), None)
+        if not clean_version:
+            source_path = gen.audio_path
+        else:
+            source_path = clean_version.audio_path
+            source_version_id = clean_version.id
+
+    if not source_path or not Path(source_path).exists():
+        raise HTTPException(status_code=404, detail="Source audio file not found")
+
+    audio, sample_rate = await asyncio.to_thread(load_audio, source_path)
+    processed_audio = await asyncio.to_thread(apply_effects, audio, sample_rate, chain_dicts)
+
+    version_id = str(uuid.uuid4())
+    processed_path = config.get_generations_dir() / f"{generation_id}_{version_id[:8]}.wav"
+    await asyncio.to_thread(save_audio, processed_audio, str(processed_path), sample_rate)
+
+    label = data.label or f"version-{len(all_versions) + 1}"
+
+    version = versions_mod.create_version(
+        generation_id=generation_id,
+        label=label,
+        audio_path=str(processed_path),
+        db=db,
+        effects_chain=chain_dicts,
+        is_default=data.set_as_default,
+        source_version_id=source_version_id,
+    )
+
+    return version
+
+
+@router.put(
+    "/generations/{generation_id}/versions/{version_id}/set-default",
+    response_model=models.GenerationVersionResponse,
+)
+async def set_default_version(
+    generation_id: str,
+    version_id: str,
+    db: Session = Depends(get_db),
+):
+    """Set a specific version as the default for a generation."""
+    from ..services import versions as versions_mod
+
+    version = versions_mod.get_version(version_id, db)
+    if not version or version.generation_id != generation_id:
+        raise HTTPException(status_code=404, detail="Version not found")
+
+    result = versions_mod.set_default_version(version_id, db)
+    if not result:
+        raise HTTPException(status_code=404, detail="Version not found")
+    return result
+
+
+@router.delete("/generations/{generation_id}/versions/{version_id}")
+async def delete_generation_version(
+    generation_id: str,
+    version_id: str,
+    db: Session = Depends(get_db),
+):
+    """Delete a version. Cannot delete the last remaining version."""
+    from ..services import versions as versions_mod
+
+    version = versions_mod.get_version(version_id, db)
+    if not version or version.generation_id != generation_id:
+        raise HTTPException(status_code=404, detail="Version not found")
+
+    if not versions_mod.delete_version(version_id, db):
+        raise HTTPException(
+            status_code=400,
+            detail="Cannot delete the last remaining version",
+        )
+    return {"status": "deleted"}
@@ -0,0 +1,276 @@
+"""TTS generation endpoints."""
+
+import asyncio
+import uuid
+
+from fastapi import APIRouter, Depends, HTTPException
+from fastapi.responses import StreamingResponse
+from sqlalchemy.orm import Session
+
+from .. import models
+from ..services import history, profiles, tts
+from ..database import Generation as DBGeneration, VoiceProfile as DBVoiceProfile, get_db
+from ..services.generation import run_generation
+from ..services.task_queue import enqueue_generation
+from ..utils.tasks import get_task_manager
+
+router = APIRouter()
+
+
+@router.post("/generate", response_model=models.GenerationResponse)
+async def generate_speech(
+    data: models.GenerationRequest,
+    db: Session = Depends(get_db),
+):
+    """Generate speech from text using a voice profile."""
+    task_manager = get_task_manager()
+    generation_id = str(uuid.uuid4())
+
+    profile = await profiles.get_profile(data.profile_id, db)
+    if not profile:
+        raise HTTPException(status_code=404, detail="Profile not found")
+
+    from ..backends import engine_has_model_sizes
+
+    engine = data.engine or "qwen"
+    model_size = (data.model_size or "1.7B") if engine_has_model_sizes(engine) else None
+
+    generation = await history.create_generation(
+        profile_id=data.profile_id,
+        text=data.text,
+        language=data.language,
+        audio_path="",
+        duration=0,
+        seed=data.seed,
+        db=db,
+        instruct=data.instruct,
+        generation_id=generation_id,
+        status="generating",
+        engine=engine,
+        model_size=model_size if engine_has_model_sizes(engine) else None,
+    )
+
+    task_manager.start_generation(
+        task_id=generation_id,
+        profile_id=data.profile_id,
+        text=data.text,
+    )
+
+    effects_chain_config = None
+    if data.effects_chain is not None:
+        effects_chain_config = [e.model_dump() for e in data.effects_chain]
+    else:
+        import json as _json
+
+        profile_obj = db.query(DBVoiceProfile).filter_by(id=data.profile_id).first()
+        if profile_obj and profile_obj.effects_chain:
+            try:
+                effects_chain_config = _json.loads(profile_obj.effects_chain)
+            except Exception:
+                pass
+
+    enqueue_generation(
+        run_generation(
+            generation_id=generation_id,
+            profile_id=data.profile_id,
+            text=data.text,
+            language=data.language,
+            engine=engine,
+            model_size=model_size,
+            seed=data.seed,
+            normalize=data.normalize,
+            effects_chain=effects_chain_config,
+            instruct=data.instruct,
+            mode="generate",
+            max_chunk_chars=data.max_chunk_chars,
+            crossfade_ms=data.crossfade_ms,
+        )
+    )
+
+    return generation
+
+
+@router.post("/generate/{generation_id}/retry", response_model=models.GenerationResponse)
+async def retry_generation(generation_id: str, db: Session = Depends(get_db)):
+    """Retry a failed generation using the same parameters."""
+    gen = db.query(DBGeneration).filter_by(id=generation_id).first()
+    if not gen:
+        raise HTTPException(status_code=404, detail="Generation not found")
+
+    if (gen.status or "completed") != "failed":
+        raise HTTPException(status_code=400, detail="Only failed generations can be retried")
+
+    gen.status = "generating"
+    gen.error = None
+    gen.audio_path = ""
+    gen.duration = 0
+    db.commit()
+    db.refresh(gen)
+
+    task_manager = get_task_manager()
+    task_manager.start_generation(
+        task_id=generation_id,
+        profile_id=gen.profile_id,
+        text=gen.text,
+    )
+
+    enqueue_generation(
+        run_generation(
+            generation_id=generation_id,
+            profile_id=gen.profile_id,
+            text=gen.text,
+            language=gen.language,
+            engine=gen.engine or "qwen",
+            model_size=gen.model_size or "1.7B",
+            seed=gen.seed,
+            instruct=gen.instruct,
+            mode="retry",
+        )
+    )
+
+    return models.GenerationResponse.model_validate(gen)
+
+
+@router.post(
+    "/generate/{generation_id}/regenerate",
+    response_model=models.GenerationResponse,
+)
+async def regenerate_generation(generation_id: str, db: Session = Depends(get_db)):
+    """Re-run TTS with the same parameters and save the result as a new version."""
+    gen = db.query(DBGeneration).filter_by(id=generation_id).first()
+    if not gen:
+        raise HTTPException(status_code=404, detail="Generation not found")
+    if (gen.status or "completed") != "completed":
+        raise HTTPException(status_code=400, detail="Generation must be completed to regenerate")
+
+    gen.status = "generating"
+    gen.error = None
+    db.commit()
+    db.refresh(gen)
+
+    task_manager = get_task_manager()
+    task_manager.start_generation(
+        task_id=generation_id,
+        profile_id=gen.profile_id,
+        text=gen.text,
+    )
+
+    version_id = str(uuid.uuid4())
+
+    enqueue_generation(
+        run_generation(
+            generation_id=generation_id,
+            profile_id=gen.profile_id,
+            text=gen.text,
+            language=gen.language,
+            engine=gen.engine or "qwen",
+            model_size=gen.model_size or "1.7B",
+            seed=gen.seed,
+            instruct=gen.instruct,
+            mode="regenerate",
+            version_id=version_id,
+        )
+    )
+
+    return models.GenerationResponse.model_validate(gen)
+
+
+@router.get("/generate/{generation_id}/status")
+async def get_generation_status(generation_id: str, db: Session = Depends(get_db)):
+    """SSE endpoint that streams generation status updates."""
+    import json
+
+    async def event_stream():
+        while True:
+            db.expire_all()
+            gen = db.query(DBGeneration).filter_by(id=generation_id).first()
+            if not gen:
+                yield f"data: {json.dumps({'status': 'not_found', 'id': generation_id})}\n\n"
+                return
+
+            payload = {
+                "id": gen.id,
+                "status": gen.status or "completed",
+                "duration": gen.duration,
+                "error": gen.error,
+            }
+            yield f"data: {json.dumps(payload)}\n\n"
+
+            if (gen.status or "completed") in ("completed", "failed"):
+                return
+
+            await asyncio.sleep(1)
+
+    return StreamingResponse(
+        event_stream(),
+        media_type="text/event-stream",
+        headers={
+            "Cache-Control": "no-cache",
+            "Connection": "keep-alive",
+            "X-Accel-Buffering": "no",
+        },
+    )
+
+
+@router.post("/generate/stream")
+async def stream_speech(
+    data: models.GenerationRequest,
+    db: Session = Depends(get_db),
+):
+    """Generate speech and stream the WAV audio directly without saving to disk."""
+    from ..backends import get_tts_backend_for_engine, ensure_model_cached_or_raise, load_engine_model, engine_needs_trim
+
+    profile = await profiles.get_profile(data.profile_id, db)
+    if not profile:
+        raise HTTPException(status_code=404, detail="Profile not found")
+
+    engine = data.engine or "qwen"
+    tts_model = get_tts_backend_for_engine(engine)
+    model_size = data.model_size or "1.7B"
+
+    await ensure_model_cached_or_raise(engine, model_size)
+    await load_engine_model(engine, model_size)
+
+    voice_prompt = await profiles.create_voice_prompt_for_profile(
+        data.profile_id,
+        db,
+        engine=engine,
+    )
+
+    from ..utils.chunked_tts import generate_chunked
+
+    trim_fn = None
+    if engine_needs_trim(engine):
+        from ..utils.audio import trim_tts_output
+
+        trim_fn = trim_tts_output
+
+    audio, sample_rate = await generate_chunked(
+        tts_model,
+        data.text,
+        voice_prompt,
+        language=data.language,
+        seed=data.seed,
+        instruct=data.instruct,
+        max_chunk_chars=data.max_chunk_chars,
+        crossfade_ms=data.crossfade_ms,
+        trim_fn=trim_fn,
+    )
+
+    if data.normalize:
+        from ..utils.audio import normalize_audio
+
+        audio = normalize_audio(audio)
+
+    wav_bytes = tts.audio_to_wav_bytes(audio, sample_rate)
+
+    async def _wav_stream():
+        chunk_size = 64 * 1024
+        for i in range(0, len(wav_bytes), chunk_size):
+            yield wav_bytes[i : i + chunk_size]
+
+    return StreamingResponse(
+        _wav_stream(),
+        media_type="audio/wav",
+        headers={"Content-Disposition": 'attachment; filename="speech.wav"'},
+    )
@@ -0,0 +1,225 @@
+"""Health and infrastructure endpoints."""
+
+import asyncio
+import os
+import signal
+
+import torch
+from fastapi import APIRouter, Depends
+from sqlalchemy.orm import Session
+
+from .. import config, models
+from ..services import tts
+from ..database import get_db
+from ..utils.platform_detect import get_backend_type
+
+router = APIRouter()
+
+
+@router.get("/")
+async def root():
+    """Root endpoint."""
+    from .. import __version__
+
+    return {"message": "voicebox API", "version": __version__}
+
+
+@router.post("/shutdown")
+async def shutdown():
+    """Gracefully shutdown the server."""
+
+    async def shutdown_async():
+        await asyncio.sleep(0.1)
+        os.kill(os.getpid(), signal.SIGTERM)
+
+    asyncio.create_task(shutdown_async())
+    return {"message": "Shutting down..."}
+
+
+@router.post("/watchdog/disable")
+async def watchdog_disable():
+    """Disable the parent process watchdog so the server keeps running."""
+    from backend.server import disable_watchdog
+
+    disable_watchdog()
+    return {"message": "Watchdog disabled"}
+
+
+@router.get("/health", response_model=models.HealthResponse)
+async def health():
+    """Health check endpoint."""
+    from huggingface_hub import constants as hf_constants
+    from pathlib import Path
+
+    tts_model = tts.get_tts_model()
+    backend_type = get_backend_type()
+
+    has_cuda = torch.cuda.is_available()
+    has_mps = hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
+
+    has_xpu = False
+    xpu_name = None
+    try:
+        import intel_extension_for_pytorch as ipex  # noqa: F401 -- side-effect import enables XPU
+
+        if hasattr(torch, "xpu") and torch.xpu.is_available():
+            has_xpu = True
+            try:
+                xpu_name = torch.xpu.get_device_name(0)
+            except Exception:
+                xpu_name = "Intel GPU"
+    except ImportError:
+        pass
+
+    has_directml = False
+    directml_name = None
+    try:
+        import torch_directml
+
+        if torch_directml.device_count() > 0:
+            has_directml = True
+            try:
+                directml_name = torch_directml.device_name(0)
+            except Exception:
+                directml_name = "DirectML GPU"
+    except ImportError:
+        pass
+
+    gpu_available = has_cuda or has_mps or has_xpu or has_directml or backend_type == "mlx"
+
+    gpu_type = None
+    if has_cuda:
+        gpu_type = f"CUDA ({torch.cuda.get_device_name(0)})"
+    elif has_mps:
+        gpu_type = "MPS (Apple Silicon)"
+    elif backend_type == "mlx":
+        gpu_type = "Metal (Apple Silicon via MLX)"
+    elif has_xpu:
+        gpu_type = f"XPU ({xpu_name})"
+    elif has_directml:
+        gpu_type = f"DirectML ({directml_name})"
+
+    vram_used = None
+    if has_cuda:
+        vram_used = torch.cuda.memory_allocated() / 1024 / 1024
+
+    model_loaded = False
+    model_size = None
+    try:
+        if tts_model.is_loaded():
+            model_loaded = True
+            model_size = getattr(tts_model, "_current_model_size", None)
+            if not model_size:
+                model_size = getattr(tts_model, "model_size", None)
+    except Exception:
+        model_loaded = False
+        model_size = None
+
+    model_downloaded = None
+    try:
+        from ..backends import get_model_config
+
+        default_config = get_model_config("qwen-tts-1.7B")
+        default_model_id = default_config.hf_repo_id if default_config else "Qwen/Qwen3-TTS-12Hz-1.7B-Base"
+
+        try:
+            from huggingface_hub import scan_cache_dir
+
+            cache_info = scan_cache_dir()
+            for repo in cache_info.repos:
+                if repo.repo_id == default_model_id:
+                    model_downloaded = True
+                    break
+        except (ImportError, Exception):
+            cache_dir = hf_constants.HF_HUB_CACHE
+            repo_cache = Path(cache_dir) / ("models--" + default_model_id.replace("/", "--"))
+            if repo_cache.exists():
+                has_model_files = (
+                    any(repo_cache.rglob("*.bin"))
+                    or any(repo_cache.rglob("*.safetensors"))
+                    or any(repo_cache.rglob("*.pt"))
+                    or any(repo_cache.rglob("*.pth"))
+                    or any(repo_cache.rglob("*.npz"))
+                )
+                model_downloaded = has_model_files
+    except Exception:
+        pass
+
+    return models.HealthResponse(
+        status="healthy",
+        model_loaded=model_loaded,
+        model_downloaded=model_downloaded,
+        model_size=model_size,
+        gpu_available=gpu_available,
+        gpu_type=gpu_type,
+        vram_used_mb=vram_used,
+        backend_type=backend_type,
+        backend_variant=os.environ.get("VOICEBOX_BACKEND_VARIANT", "cuda" if torch.cuda.is_available() else "cpu"),
+    )
+
+
+@router.get("/health/filesystem", response_model=models.FilesystemHealthResponse)
+async def filesystem_health():
+    """Check filesystem health: directory existence, write permissions, and disk space."""
+    import shutil
+
+    dirs_to_check = {
+        "generations": config.get_generations_dir(),
+        "profiles": config.get_profiles_dir(),
+        "data": config.get_data_dir(),
+    }
+
+    checks: list[models.DirectoryCheck] = []
+    all_ok = True
+
+    for _label, dir_path in dirs_to_check.items():
+        exists = dir_path.exists()
+        writable = False
+        error = None
+        if exists:
+            probe = dir_path / ".voicebox_probe"
+            try:
+                probe.write_text("ok")
+                probe.unlink()
+                writable = True
+            except PermissionError:
+                error = "Permission denied"
+            except OSError as e:
+                error = str(e)
+            finally:
+                try:
+                    probe.unlink(missing_ok=True)
+                except Exception:
+                    pass
+        else:
+            error = "Directory does not exist"
+
+        if not exists or not writable:
+            all_ok = False
+
+        checks.append(
+            models.DirectoryCheck(
+                path=str(dir_path),
+                exists=exists,
+                writable=writable,
+                error=error,
+            )
+        )
+
+    disk_free_mb = None
+    disk_total_mb = None
+    try:
+        usage = shutil.disk_usage(str(config.get_data_dir()))
+        disk_free_mb = round(usage.free / (1024 * 1024), 1)
+        disk_total_mb = round(usage.total / (1024 * 1024), 1)
+        if disk_free_mb < 500:
+            all_ok = False
+    except OSError:
+        all_ok = False
+
+    return models.FilesystemHealthResponse(
+        healthy=all_ok,
+        disk_free_mb=disk_free_mb,
+        disk_total_mb=disk_total_mb,
+        directories=checks,
+    )
@@ -0,0 +1,178 @@
+"""Generation history endpoints."""
+
+import io
+from pathlib import Path
+
+from fastapi import APIRouter, Depends, File, HTTPException, UploadFile
+from fastapi.responses import FileResponse, StreamingResponse
+from sqlalchemy.orm import Session
+
+from .. import models
+from ..services import export_import, history
+from ..app import safe_content_disposition
+from ..database import Generation as DBGeneration, VoiceProfile as DBVoiceProfile, get_db
+
+router = APIRouter()
+
+
+@router.get("/history", response_model=models.HistoryListResponse)
+async def list_history(
+    profile_id: str | None = None,
+    search: str | None = None,
+    limit: int = 50,
+    offset: int = 0,
+    db: Session = Depends(get_db),
+):
+    """List generation history with optional filters."""
+    query = models.HistoryQuery(
+        profile_id=profile_id,
+        search=search,
+        limit=limit,
+        offset=offset,
+    )
+    return await history.list_generations(query, db)
+
+
+@router.get("/history/stats")
+async def get_stats(db: Session = Depends(get_db)):
+    """Get generation statistics."""
+    return await history.get_generation_stats(db)
+
+
+@router.post("/history/import")
+async def import_generation(
+    file: UploadFile = File(...),
+    db: Session = Depends(get_db),
+):
+    """Import a generation from a ZIP archive."""
+    MAX_FILE_SIZE = 50 * 1024 * 1024
+
+    content = await file.read()
+
+    if len(content) > MAX_FILE_SIZE:
+        raise HTTPException(
+            status_code=400, detail=f"File too large. Maximum size is {MAX_FILE_SIZE / (1024 * 1024)}MB"
+        )
+
+    try:
+        result = await export_import.import_generation_from_zip(content, db)
+        return result
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@router.get("/history/{generation_id}", response_model=models.HistoryResponse)
+async def get_generation(
+    generation_id: str,
+    db: Session = Depends(get_db),
+):
+    """Get a generation by ID."""
+    result = (
+        db.query(DBGeneration, DBVoiceProfile.name.label("profile_name"))
+        .join(DBVoiceProfile, DBGeneration.profile_id == DBVoiceProfile.id)
+        .filter(DBGeneration.id == generation_id)
+        .first()
+    )
+
+    if not result:
+        raise HTTPException(status_code=404, detail="Generation not found")
+
+    gen, profile_name = result
+    return models.HistoryResponse(
+        id=gen.id,
+        profile_id=gen.profile_id,
+        profile_name=profile_name,
+        text=gen.text,
+        language=gen.language,
+        audio_path=gen.audio_path,
+        duration=gen.duration,
+        seed=gen.seed,
+        instruct=gen.instruct,
+        created_at=gen.created_at,
+    )
+
+
+@router.post("/history/{generation_id}/favorite")
+async def toggle_favorite(
+    generation_id: str,
+    db: Session = Depends(get_db),
+):
+    """Toggle the favorite status of a generation."""
+    gen = db.query(DBGeneration).filter_by(id=generation_id).first()
+    if not gen:
+        raise HTTPException(status_code=404, detail="Generation not found")
+    gen.is_favorited = not gen.is_favorited
+    db.commit()
+    return {"is_favorited": gen.is_favorited}
+
+
+@router.delete("/history/{generation_id}")
+async def delete_generation(
+    generation_id: str,
+    db: Session = Depends(get_db),
+):
+    """Delete a generation."""
+    success = await history.delete_generation(generation_id, db)
+    if not success:
+        raise HTTPException(status_code=404, detail="Generation not found")
+    return {"message": "Generation deleted successfully"}
+
+
+@router.get("/history/{generation_id}/export")
+async def export_generation(
+    generation_id: str,
+    db: Session = Depends(get_db),
+):
+    """Export a generation as a ZIP archive."""
+    generation = db.query(DBGeneration).filter_by(id=generation_id).first()
+    if not generation:
+        raise HTTPException(status_code=404, detail="Generation not found")
+
+    try:
+        zip_bytes = export_import.export_generation_to_zip(generation_id, db)
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+    safe_text = "".join(c for c in generation.text[:30] if c.isalnum() or c in (" ", "-", "_")).strip()
+    if not safe_text:
+        safe_text = "generation"
+    filename = f"generation-{safe_text}.voicebox.zip"
+
+    return StreamingResponse(
+        io.BytesIO(zip_bytes),
+        media_type="application/zip",
+        headers={"Content-Disposition": safe_content_disposition("attachment", filename)},
+    )
+
+
+@router.get("/history/{generation_id}/export-audio")
+async def export_generation_audio(
+    generation_id: str,
+    db: Session = Depends(get_db),
+):
+    """Export only the audio file from a generation."""
+    generation = db.query(DBGeneration).filter_by(id=generation_id).first()
+    if not generation:
+        raise HTTPException(status_code=404, detail="Generation not found")
+
+    if not generation.audio_path:
+        raise HTTPException(status_code=404, detail="Generation has no audio file")
+
+    audio_path = Path(generation.audio_path)
+    if not audio_path.is_file():
+        raise HTTPException(status_code=404, detail="Audio file not found")
+
+    safe_text = "".join(c for c in generation.text[:30] if c.isalnum() or c in (" ", "-", "_")).strip()
+    if not safe_text:
+        safe_text = "generation"
+    filename = f"{safe_text}.wav"
+
+    return FileResponse(
+        audio_path,
+        media_type="audio/wav",
+        headers={"Content-Disposition": safe_content_disposition("attachment", filename)},
+    )
@@ -0,0 +1,474 @@
+"""Model management endpoints."""
+
+import asyncio
+import shutil
+from pathlib import Path
+
+from fastapi import APIRouter, Depends, HTTPException
+from fastapi.responses import StreamingResponse
+from sqlalchemy.orm import Session
+
+from .. import models
+from ..utils.platform_detect import get_backend_type
+from ..services.task_queue import create_background_task
+from ..utils.progress import get_progress_manager
+from ..utils.tasks import get_task_manager
+
+router = APIRouter()
+
+
+def _get_dir_size(path: Path) -> int:
+    """Get total size of a directory in bytes."""
+    total = 0
+    for f in path.rglob("*"):
+        if f.is_file():
+            total += f.stat().st_size
+    return total
+
+
+def _copy_with_progress(src: Path, dst: Path, progress_manager, copied_so_far: int, total_bytes: int) -> int:
+    """Copy a directory tree with byte-level progress tracking."""
+    dst.mkdir(parents=True, exist_ok=True)
+    for item in src.iterdir():
+        dest_item = dst / item.name
+        if item.is_dir():
+            copied_so_far = _copy_with_progress(item, dest_item, progress_manager, copied_so_far, total_bytes)
+        else:
+            size = item.stat().st_size
+            shutil.copy2(str(item), str(dest_item))
+            copied_so_far += size
+            progress_manager.update_progress(
+                "migration",
+                copied_so_far,
+                total_bytes,
+                filename=item.name,
+                status="downloading",
+            )
+    return copied_so_far
+
+
+@router.post("/models/load")
+async def load_model(model_size: str = "1.7B"):
+    """Manually load TTS model."""
+    from ..services import tts
+
+    try:
+        tts_model = tts.get_tts_model()
+        await tts_model.load_model_async(model_size)
+        return {"message": f"Model {model_size} loaded successfully"}
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@router.post("/models/unload")
+async def unload_model():
+    """Unload the default Qwen TTS model to free memory."""
+    from ..services import tts
+
+    try:
+        tts.unload_tts_model()
+        return {"message": "Model unloaded successfully"}
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@router.post("/models/{model_name}/unload")
+async def unload_model_by_name(model_name: str):
+    """Unload a specific model from memory without deleting it from disk."""
+    from ..backends import get_model_config, unload_model_by_config
+
+    config = get_model_config(model_name)
+    if not config:
+        raise HTTPException(status_code=400, detail=f"Unknown model: {model_name}")
+
+    try:
+        was_loaded = unload_model_by_config(config)
+        if not was_loaded:
+            return {"message": f"Model {model_name} is not loaded"}
+        return {"message": f"Model {model_name} unloaded successfully"}
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e)) from e
+
+
+@router.get("/models/progress/{model_name}")
+async def get_model_progress(model_name: str):
+    """Get model download progress via Server-Sent Events."""
+    progress_manager = get_progress_manager()
+
+    async def event_generator():
+        async for event in progress_manager.subscribe(model_name):
+            yield event
+
+    return StreamingResponse(
+        event_generator(),
+        media_type="text/event-stream",
+        headers={
+            "Cache-Control": "no-cache",
+            "Connection": "keep-alive",
+            "X-Accel-Buffering": "no",
+        },
+    )
+
+
+@router.get("/models/cache-dir")
+async def get_models_cache_dir():
+    """Get the path to the HuggingFace model cache directory."""
+    from huggingface_hub import constants as hf_constants
+
+    return {"path": str(Path(hf_constants.HF_HUB_CACHE))}
+
+
+@router.post("/models/migrate")
+async def migrate_models(request: models.ModelMigrateRequest):
+    """Move all downloaded models to a new directory with byte-level progress via SSE."""
+    from huggingface_hub import constants as hf_constants
+
+    source = Path(hf_constants.HF_HUB_CACHE)
+    destination = Path(request.destination)
+
+    if not source.exists():
+        raise HTTPException(status_code=404, detail="Current model cache directory not found")
+
+    if source.resolve() == destination.resolve():
+        raise HTTPException(status_code=400, detail="Source and destination are the same directory")
+
+    if destination.resolve().is_relative_to(source.resolve()):
+        raise HTTPException(status_code=400, detail="Destination cannot be inside the current cache directory")
+
+    model_dirs = [d for d in source.iterdir() if d.name.startswith("models--") and d.is_dir()]
+    if not model_dirs:
+        return {"moved": 0, "errors": [], "source": str(source), "destination": str(destination)}
+
+    destination.mkdir(parents=True, exist_ok=True)
+
+    progress_manager = get_progress_manager()
+
+    same_fs = False
+    try:
+        same_fs = source.stat().st_dev == destination.stat().st_dev
+    except OSError:
+        pass
+
+    async def migrate_background():
+        moved = 0
+        errors = []
+        try:
+            if same_fs:
+                total = len(model_dirs)
+                for i, item in enumerate(model_dirs):
+                    dest_item = destination / item.name
+                    try:
+                        if dest_item.exists():
+                            shutil.rmtree(dest_item)
+                        shutil.move(str(item), str(dest_item))
+                        moved += 1
+                        progress_manager.update_progress(
+                            "migration",
+                            i + 1,
+                            total,
+                            filename=item.name,
+                            status="downloading",
+                        )
+                    except Exception as e:
+                        errors.append(f"{item.name}: {str(e)}")
+            else:
+                total_bytes = sum(_get_dir_size(d) for d in model_dirs)
+                progress_manager.update_progress(
+                    "migration", 0, total_bytes, filename="Calculating...", status="downloading"
+                )
+
+                copied = 0
+                for item in model_dirs:
+                    dest_item = destination / item.name
+                    try:
+                        if dest_item.exists():
+                            shutil.rmtree(dest_item)
+                        copied = await asyncio.to_thread(
+                            _copy_with_progress, item, dest_item, progress_manager, copied, total_bytes
+                        )
+                        await asyncio.to_thread(shutil.rmtree, str(item))
+                        moved += 1
+                    except Exception as e:
+                        errors.append(f"{item.name}: {str(e)}")
+
+            progress_manager.update_progress("migration", 1, 1, status="complete")
+            progress_manager.mark_complete("migration")
+        except Exception as e:
+            progress_manager.update_progress("migration", 0, 0, status="error")
+            progress_manager.mark_error("migration", str(e))
+
+    create_background_task(migrate_background())
+
+    return {"source": str(source), "destination": str(destination)}
+
+
+@router.get("/models/migrate/progress")
+async def get_migration_progress():
+    """Get model migration progress via Server-Sent Events."""
+    progress_manager = get_progress_manager()
+
+    async def event_generator():
+        async for event in progress_manager.subscribe("migration"):
+            yield event
+
+    return StreamingResponse(
+        event_generator(),
+        media_type="text/event-stream",
+        headers={
+            "Cache-Control": "no-cache",
+            "Connection": "keep-alive",
+            "X-Accel-Buffering": "no",
+        },
+    )
+
+
+@router.get("/models/status", response_model=models.ModelStatusListResponse)
+async def get_model_status():
+    """Get status of all available models."""
+    from huggingface_hub import constants as hf_constants
+
+    backend_type = get_backend_type()
+    task_manager = get_task_manager()
+
+    active_download_names = {task.model_name for task in task_manager.get_active_downloads()}
+
+    try:
+        from huggingface_hub import scan_cache_dir
+
+        use_scan_cache = True
+    except ImportError:
+        use_scan_cache = False
+
+    from ..backends import get_all_model_configs, check_model_loaded
+
+    registry_configs = get_all_model_configs()
+    model_configs = [
+        {
+            "model_name": cfg.model_name,
+            "display_name": cfg.display_name,
+            "hf_repo_id": cfg.hf_repo_id,
+            "model_size": cfg.model_size,
+            "check_loaded": lambda c=cfg: check_model_loaded(c),
+        }
+        for cfg in registry_configs
+    ]
+
+    model_to_repo = {cfg["model_name"]: cfg["hf_repo_id"] for cfg in model_configs}
+    active_download_repos = {model_to_repo.get(name) for name in active_download_names if name in model_to_repo}
+
+    cache_info = None
+    if use_scan_cache:
+        try:
+            cache_info = scan_cache_dir()
+        except Exception:
+            pass
+
+    statuses = []
+
+    for config in model_configs:
+        try:
+            downloaded = False
+            size_mb = None
+            loaded = False
+
+            if cache_info:
+                repo_id = config["hf_repo_id"]
+                for repo in cache_info.repos:
+                    if repo.repo_id == repo_id:
+                        has_model_weights = False
+                        for rev in repo.revisions:
+                            for f in rev.files:
+                                fname = f.file_name.lower()
+                                if fname.endswith((".safetensors", ".bin", ".pt", ".pth", ".npz")):
+                                    has_model_weights = True
+                                    break
+                            if has_model_weights:
+                                break
+
+                        has_incomplete = False
+                        try:
+                            cache_dir = hf_constants.HF_HUB_CACHE
+                            blobs_dir = Path(cache_dir) / ("models--" + repo_id.replace("/", "--")) / "blobs"
+                            if blobs_dir.exists():
+                                has_incomplete = any(blobs_dir.glob("*.incomplete"))
+                        except Exception:
+                            pass
+
+                        if has_model_weights and not has_incomplete:
+                            downloaded = True
+                            try:
+                                total_size = sum(revision.size_on_disk for revision in repo.revisions)
+                                size_mb = total_size / (1024 * 1024)
+                            except Exception:
+                                pass
+                        break
+
+            if not downloaded:
+                try:
+                    cache_dir = hf_constants.HF_HUB_CACHE
+                    repo_cache = Path(cache_dir) / ("models--" + config["hf_repo_id"].replace("/", "--"))
+
+                    if repo_cache.exists():
+                        blobs_dir = repo_cache / "blobs"
+                        has_incomplete = blobs_dir.exists() and any(blobs_dir.glob("*.incomplete"))
+
+                        if not has_incomplete:
+                            snapshots_dir = repo_cache / "snapshots"
+                            has_model_files = False
+                            if snapshots_dir.exists():
+                                has_model_files = (
+                                    any(snapshots_dir.rglob("*.bin"))
+                                    or any(snapshots_dir.rglob("*.safetensors"))
+                                    or any(snapshots_dir.rglob("*.pt"))
+                                    or any(snapshots_dir.rglob("*.pth"))
+                                    or any(snapshots_dir.rglob("*.npz"))
+                                )
+
+                            if has_model_files:
+                                downloaded = True
+                                try:
+                                    total_size = sum(
+                                        f.stat().st_size
+                                        for f in repo_cache.rglob("*")
+                                        if f.is_file() and not f.name.endswith(".incomplete")
+                                    )
+                                    size_mb = total_size / (1024 * 1024)
+                                except Exception:
+                                    pass
+                except Exception:
+                    pass
+
+            try:
+                loaded = config["check_loaded"]()
+            except Exception:
+                loaded = False
+
+            is_downloading = config["hf_repo_id"] in active_download_repos
+
+            if is_downloading:
+                downloaded = False
+                size_mb = None
+
+            statuses.append(
+                models.ModelStatus(
+                    model_name=config["model_name"],
+                    display_name=config["display_name"],
+                    hf_repo_id=config["hf_repo_id"],
+                    downloaded=downloaded,
+                    downloading=is_downloading,
+                    size_mb=size_mb,
+                    loaded=loaded,
+                )
+            )
+        except Exception:
+            try:
+                loaded = config["check_loaded"]()
+            except Exception:
+                loaded = False
+
+            is_downloading = config["hf_repo_id"] in active_download_repos
+
+            statuses.append(
+                models.ModelStatus(
+                    model_name=config["model_name"],
+                    display_name=config["display_name"],
+                    hf_repo_id=config["hf_repo_id"],
+                    downloaded=False,
+                    downloading=is_downloading,
+                    size_mb=None,
+                    loaded=loaded,
+                )
+            )
+
+    return models.ModelStatusListResponse(models=statuses)
+
+
+@router.post("/models/download")
+async def trigger_model_download(request: models.ModelDownloadRequest):
+    """Trigger download of a specific model."""
+    from ..backends import get_model_config, get_model_load_func
+
+    task_manager = get_task_manager()
+    progress_manager = get_progress_manager()
+
+    config = get_model_config(request.model_name)
+    if not config:
+        raise HTTPException(status_code=400, detail=f"Unknown model: {request.model_name}")
+
+    load_func = get_model_load_func(config)
+
+    async def download_in_background():
+        try:
+            result = load_func()
+            if asyncio.iscoroutine(result):
+                await result
+            task_manager.complete_download(request.model_name)
+        except Exception as e:
+            task_manager.error_download(request.model_name, str(e))
+
+    task_manager.start_download(request.model_name)
+
+    progress_manager.update_progress(
+        model_name=request.model_name,
+        current=0,
+        total=0,
+        filename="Connecting to HuggingFace...",
+        status="downloading",
+    )
+
+    create_background_task(download_in_background())
+
+    return {"message": f"Model {request.model_name} download started"}
+
+
+@router.post("/models/download/cancel")
+async def cancel_model_download(request: models.ModelDownloadRequest):
+    """Cancel or dismiss an errored/stale download task."""
+    task_manager = get_task_manager()
+    progress_manager = get_progress_manager()
+
+    removed = task_manager.cancel_download(request.model_name)
+
+    progress_removed = False
+    with progress_manager._lock:
+        if request.model_name in progress_manager._progress:
+            del progress_manager._progress[request.model_name]
+            progress_removed = True
+
+    if removed or progress_removed:
+        return {"message": f"Download task for {request.model_name} cancelled"}
+    return {"message": f"No active task found for {request.model_name}"}
+
+
+@router.delete("/models/{model_name}")
+async def delete_model(model_name: str):
+    """Delete a downloaded model from the HuggingFace cache."""
+    from huggingface_hub import constants as hf_constants
+    from ..backends import get_model_config, unload_model_by_config
+
+    config = get_model_config(model_name)
+    if not config:
+        raise HTTPException(status_code=400, detail=f"Unknown model: {model_name}")
+
+    hf_repo_id = config.hf_repo_id
+
+    try:
+        unload_model_by_config(config)
+
+        cache_dir = hf_constants.HF_HUB_CACHE
+        repo_cache_dir = Path(cache_dir) / ("models--" + hf_repo_id.replace("/", "--"))
+
+        if not repo_cache_dir.exists():
+            raise HTTPException(status_code=404, detail=f"Model {model_name} not found in cache")
+
+        try:
+            shutil.rmtree(repo_cache_dir)
+        except OSError as e:
+            raise HTTPException(status_code=500, detail=f"Failed to delete model cache directory: {str(e)}")
+
+        return {"message": f"Model {model_name} deleted successfully"}
+
+    except HTTPException:
+        raise
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Failed to delete model: {str(e)}")
@@ -0,0 +1,309 @@
+"""Voice profile endpoints."""
+
+import io
+import tempfile
+from datetime import datetime
+from pathlib import Path
+
+from fastapi import APIRouter, Depends, File, Form, HTTPException, UploadFile
+from fastapi.responses import FileResponse, StreamingResponse
+from sqlalchemy.orm import Session
+
+from .. import config, models
+from ..app import safe_content_disposition
+from ..database import VoiceProfile as DBVoiceProfile, get_db
+from ..services import channels, export_import, profiles
+from ..services.profiles import _profile_to_response
+
+router = APIRouter()
+
+
+@router.post("/profiles", response_model=models.VoiceProfileResponse)
+async def create_profile(
+    data: models.VoiceProfileCreate,
+    db: Session = Depends(get_db),
+):
+    """Create a new voice profile."""
+    try:
+        return await profiles.create_profile(data, db)
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+    except Exception as e:
+        raise HTTPException(status_code=400, detail=str(e))
+
+
+@router.get("/profiles", response_model=list[models.VoiceProfileResponse])
+async def list_profiles(db: Session = Depends(get_db)):
+    """List all voice profiles."""
+    return await profiles.list_profiles(db)
+
+
+@router.post("/profiles/import", response_model=models.VoiceProfileResponse)
+async def import_profile(
+    file: UploadFile = File(...),
+    db: Session = Depends(get_db),
+):
+    """Import a voice profile from a ZIP archive."""
+    MAX_FILE_SIZE = 100 * 1024 * 1024
+
+    content = await file.read()
+
+    if len(content) > MAX_FILE_SIZE:
+        raise HTTPException(
+            status_code=400, detail=f"File too large. Maximum size is {MAX_FILE_SIZE / (1024 * 1024)}MB"
+        )
+
+    try:
+        profile = await export_import.import_profile_from_zip(content, db)
+        return profile
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@router.get("/profiles/{profile_id}", response_model=models.VoiceProfileResponse)
+async def get_profile(
+    profile_id: str,
+    db: Session = Depends(get_db),
+):
+    """Get a voice profile by ID."""
+    profile = await profiles.get_profile(profile_id, db)
+    if not profile:
+        raise HTTPException(status_code=404, detail="Profile not found")
+    return profile
+
+
+@router.put("/profiles/{profile_id}", response_model=models.VoiceProfileResponse)
+async def update_profile(
+    profile_id: str,
+    data: models.VoiceProfileCreate,
+    db: Session = Depends(get_db),
+):
+    """Update a voice profile."""
+    try:
+        profile = await profiles.update_profile(profile_id, data, db)
+        if not profile:
+            raise HTTPException(status_code=404, detail="Profile not found")
+        return profile
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+
+
+@router.delete("/profiles/{profile_id}")
+async def delete_profile(
+    profile_id: str,
+    db: Session = Depends(get_db),
+):
+    """Delete a voice profile."""
+    success = await profiles.delete_profile(profile_id, db)
+    if not success:
+        raise HTTPException(status_code=404, detail="Profile not found")
+    return {"message": "Profile deleted successfully"}
+
+
+@router.post("/profiles/{profile_id}/samples", response_model=models.ProfileSampleResponse)
+async def add_profile_sample(
+    profile_id: str,
+    file: UploadFile = File(...),
+    reference_text: str = Form(...),
+    db: Session = Depends(get_db),
+):
+    """Add a sample to a voice profile."""
+    _allowed_audio_exts = {".wav", ".mp3", ".m4a", ".ogg", ".flac", ".aac", ".webm", ".opus"}
+    _uploaded_ext = Path(file.filename or "").suffix.lower()
+    file_suffix = _uploaded_ext if _uploaded_ext in _allowed_audio_exts else ".wav"
+
+    with tempfile.NamedTemporaryFile(suffix=file_suffix, delete=False) as tmp:
+        content = await file.read()
+        tmp.write(content)
+        tmp_path = tmp.name
+
+    try:
+        sample = await profiles.add_profile_sample(
+            profile_id,
+            tmp_path,
+            reference_text,
+            db,
+        )
+        return sample
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Failed to process audio file: {str(e)}")
+    finally:
+        Path(tmp_path).unlink(missing_ok=True)
+
+
+@router.get("/profiles/{profile_id}/samples", response_model=list[models.ProfileSampleResponse])
+async def get_profile_samples(
+    profile_id: str,
+    db: Session = Depends(get_db),
+):
+    """Get all samples for a profile."""
+    return await profiles.get_profile_samples(profile_id, db)
+
+
+@router.delete("/profiles/samples/{sample_id}")
+async def delete_profile_sample(
+    sample_id: str,
+    db: Session = Depends(get_db),
+):
+    """Delete a profile sample."""
+    success = await profiles.delete_profile_sample(sample_id, db)
+    if not success:
+        raise HTTPException(status_code=404, detail="Sample not found")
+    return {"message": "Sample deleted successfully"}
+
+
+@router.put("/profiles/samples/{sample_id}", response_model=models.ProfileSampleResponse)
+async def update_profile_sample(
+    sample_id: str,
+    data: models.ProfileSampleUpdate,
+    db: Session = Depends(get_db),
+):
+    """Update a profile sample's reference text."""
+    sample = await profiles.update_profile_sample(sample_id, data.reference_text, db)
+    if not sample:
+        raise HTTPException(status_code=404, detail="Sample not found")
+    return sample
+
+
+@router.post("/profiles/{profile_id}/avatar", response_model=models.VoiceProfileResponse)
+async def upload_profile_avatar(
+    profile_id: str,
+    file: UploadFile = File(...),
+    db: Session = Depends(get_db),
+):
+    """Upload or update avatar image for a profile."""
+    with tempfile.NamedTemporaryFile(delete=False, suffix=Path(file.filename).suffix) as tmp:
+        content = await file.read()
+        tmp.write(content)
+        tmp_path = tmp.name
+
+    try:
+        profile = await profiles.upload_avatar(profile_id, tmp_path, db)
+        return profile
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+    finally:
+        Path(tmp_path).unlink(missing_ok=True)
+
+
+@router.get("/profiles/{profile_id}/avatar")
+async def get_profile_avatar(
+    profile_id: str,
+    db: Session = Depends(get_db),
+):
+    """Get avatar image for a profile."""
+    profile = await profiles.get_profile(profile_id, db)
+    if not profile:
+        raise HTTPException(status_code=404, detail="Profile not found")
+
+    if not profile.avatar_path:
+        raise HTTPException(status_code=404, detail="No avatar found for this profile")
+
+    avatar_path = Path(profile.avatar_path)
+    if not avatar_path.exists():
+        raise HTTPException(status_code=404, detail="Avatar file not found")
+
+    return FileResponse(avatar_path)
+
+
+@router.delete("/profiles/{profile_id}/avatar")
+async def delete_profile_avatar(
+    profile_id: str,
+    db: Session = Depends(get_db),
+):
+    """Delete avatar image for a profile."""
+    success = await profiles.delete_avatar(profile_id, db)
+    if not success:
+        raise HTTPException(status_code=404, detail="Profile not found or no avatar to delete")
+    return {"message": "Avatar deleted successfully"}
+
+
+@router.get("/profiles/{profile_id}/export")
+async def export_profile(
+    profile_id: str,
+    db: Session = Depends(get_db),
+):
+    """Export a voice profile as a ZIP archive."""
+    try:
+        profile = await profiles.get_profile(profile_id, db)
+        if not profile:
+            raise HTTPException(status_code=404, detail="Profile not found")
+
+        zip_bytes = export_import.export_profile_to_zip(profile_id, db)
+
+        safe_name = "".join(c for c in profile.name if c.isalnum() or c in (" ", "-", "_")).strip()
+        if not safe_name:
+            safe_name = "profile"
+        filename = f"profile-{safe_name}.voicebox.zip"
+
+        return StreamingResponse(
+            io.BytesIO(zip_bytes),
+            media_type="application/zip",
+            headers={"Content-Disposition": safe_content_disposition("attachment", filename)},
+        )
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@router.get("/profiles/{profile_id}/channels")
+async def get_profile_channels(
+    profile_id: str,
+    db: Session = Depends(get_db),
+):
+    """Get list of channel IDs assigned to a profile."""
+    try:
+        channel_ids = await channels.get_profile_channels(profile_id, db)
+        return {"channel_ids": channel_ids}
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+
+
+@router.put("/profiles/{profile_id}/channels")
+async def set_profile_channels(
+    profile_id: str,
+    data: models.ProfileChannelAssignment,
+    db: Session = Depends(get_db),
+):
+    """Set which channels a profile is assigned to."""
+    try:
+        await channels.set_profile_channels(profile_id, data, db)
+        return {"message": "Profile channels updated successfully"}
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+
+
+@router.put("/profiles/{profile_id}/effects", response_model=models.VoiceProfileResponse)
+async def update_profile_effects(
+    profile_id: str,
+    data: models.ProfileEffectsUpdate,
+    db: Session = Depends(get_db),
+):
+    """Set or clear the default effects chain for a voice profile."""
+    import json as _json
+
+    profile = db.query(DBVoiceProfile).filter_by(id=profile_id).first()
+    if not profile:
+        raise HTTPException(status_code=404, detail="Profile not found")
+
+    if data.effects_chain is not None:
+        from ..utils.effects import validate_effects_chain
+
+        chain_dicts = [e.model_dump() for e in data.effects_chain]
+        error = validate_effects_chain(chain_dicts)
+        if error:
+            raise HTTPException(status_code=400, detail=error)
+        profile.effects_chain = _json.dumps(chain_dicts)
+    else:
+        profile.effects_chain = None
+
+    profile.updated_at = datetime.utcnow()
+    db.commit()
+    db.refresh(profile)
+
+    return _profile_to_response(profile)
@@ -0,0 +1,223 @@
+"""Story endpoints."""
+
+import io
+
+from fastapi import APIRouter, Depends, HTTPException
+from fastapi.responses import StreamingResponse
+from sqlalchemy.orm import Session
+
+from .. import database, models
+from ..services import stories
+from ..app import safe_content_disposition
+from ..database import get_db
+
+router = APIRouter()
+
+
+@router.get("/stories", response_model=list[models.StoryResponse])
+async def list_stories(db: Session = Depends(get_db)):
+    """List all stories."""
+    return await stories.list_stories(db)
+
+
+@router.post("/stories", response_model=models.StoryResponse)
+async def create_story(
+    data: models.StoryCreate,
+    db: Session = Depends(get_db),
+):
+    """Create a new story."""
+    try:
+        return await stories.create_story(data, db)
+    except Exception as e:
+        raise HTTPException(status_code=400, detail=str(e))
+
+
+@router.get("/stories/{story_id}", response_model=models.StoryDetailResponse)
+async def get_story(
+    story_id: str,
+    db: Session = Depends(get_db),
+):
+    """Get a story with all its items."""
+    story = await stories.get_story(story_id, db)
+    if not story:
+        raise HTTPException(status_code=404, detail="Story not found")
+    return story
+
+
+@router.put("/stories/{story_id}", response_model=models.StoryResponse)
+async def update_story(
+    story_id: str,
+    data: models.StoryCreate,
+    db: Session = Depends(get_db),
+):
+    """Update a story."""
+    story = await stories.update_story(story_id, data, db)
+    if not story:
+        raise HTTPException(status_code=404, detail="Story not found")
+    return story
+
+
+@router.delete("/stories/{story_id}")
+async def delete_story(
+    story_id: str,
+    db: Session = Depends(get_db),
+):
+    """Delete a story."""
+    success = await stories.delete_story(story_id, db)
+    if not success:
+        raise HTTPException(status_code=404, detail="Story not found")
+    return {"message": "Story deleted successfully"}
+
+
+@router.post("/stories/{story_id}/items", response_model=models.StoryItemDetail)
+async def add_story_item(
+    story_id: str,
+    data: models.StoryItemCreate,
+    db: Session = Depends(get_db),
+):
+    """Add a generation to a story."""
+    item = await stories.add_item_to_story(story_id, data, db)
+    if not item:
+        raise HTTPException(status_code=404, detail="Story or generation not found")
+    return item
+
+
+@router.delete("/stories/{story_id}/items/{item_id}")
+async def remove_story_item(
+    story_id: str,
+    item_id: str,
+    db: Session = Depends(get_db),
+):
+    """Remove a story item from a story."""
+    success = await stories.remove_item_from_story(story_id, item_id, db)
+    if not success:
+        raise HTTPException(status_code=404, detail="Story item not found")
+    return {"message": "Item removed successfully"}
+
+
+@router.put("/stories/{story_id}/items/times")
+async def update_story_item_times(
+    story_id: str,
+    data: models.StoryItemBatchUpdate,
+    db: Session = Depends(get_db),
+):
+    """Update story item timecodes."""
+    success = await stories.update_story_item_times(story_id, data, db)
+    if not success:
+        raise HTTPException(status_code=400, detail="Invalid timecode update request")
+    return {"message": "Item timecodes updated successfully"}
+
+
+@router.put("/stories/{story_id}/items/reorder", response_model=list[models.StoryItemDetail])
+async def reorder_story_items(
+    story_id: str,
+    data: models.StoryItemReorder,
+    db: Session = Depends(get_db),
+):
+    """Reorder story items and recalculate timecodes."""
+    items = await stories.reorder_story_items(story_id, data.generation_ids, db)
+    if items is None:
+        raise HTTPException(
+            status_code=400, detail="Invalid reorder request - ensure all generation IDs belong to this story"
+        )
+    return items
+
+
+@router.put("/stories/{story_id}/items/{item_id}/move", response_model=models.StoryItemDetail)
+async def move_story_item(
+    story_id: str,
+    item_id: str,
+    data: models.StoryItemMove,
+    db: Session = Depends(get_db),
+):
+    """Move a story item (update position and/or track)."""
+    item = await stories.move_story_item(story_id, item_id, data, db)
+    if item is None:
+        raise HTTPException(status_code=404, detail="Story item not found")
+    return item
+
+
+@router.put("/stories/{story_id}/items/{item_id}/trim", response_model=models.StoryItemDetail)
+async def trim_story_item(
+    story_id: str,
+    item_id: str,
+    data: models.StoryItemTrim,
+    db: Session = Depends(get_db),
+):
+    """Trim a story item."""
+    item = await stories.trim_story_item(story_id, item_id, data, db)
+    if item is None:
+        raise HTTPException(status_code=404, detail="Story item not found or invalid trim values")
+    return item
+
+
+@router.post("/stories/{story_id}/items/{item_id}/split", response_model=list[models.StoryItemDetail])
+async def split_story_item(
+    story_id: str,
+    item_id: str,
+    data: models.StoryItemSplit,
+    db: Session = Depends(get_db),
+):
+    """Split a story item at a given time, creating two clips."""
+    items = await stories.split_story_item(story_id, item_id, data, db)
+    if items is None:
+        raise HTTPException(status_code=404, detail="Story item not found or invalid split point")
+    return items
+
+
+@router.post("/stories/{story_id}/items/{item_id}/duplicate", response_model=models.StoryItemDetail)
+async def duplicate_story_item(
+    story_id: str,
+    item_id: str,
+    db: Session = Depends(get_db),
+):
+    """Duplicate a story item."""
+    item = await stories.duplicate_story_item(story_id, item_id, db)
+    if item is None:
+        raise HTTPException(status_code=404, detail="Story item not found")
+    return item
+
+
+@router.put("/stories/{story_id}/items/{item_id}/version", response_model=models.StoryItemDetail)
+async def set_story_item_version(
+    story_id: str,
+    item_id: str,
+    data: models.StoryItemVersionUpdate,
+    db: Session = Depends(get_db),
+):
+    """Pin a story item to a specific generation version."""
+    item = await stories.set_story_item_version(story_id, item_id, data, db)
+    if item is None:
+        raise HTTPException(status_code=404, detail="Story item or version not found")
+    return item
+
+
+@router.get("/stories/{story_id}/export-audio")
+async def export_story_audio(
+    story_id: str,
+    db: Session = Depends(get_db),
+):
+    """Export story as single mixed audio file."""
+    try:
+        story = db.query(database.Story).filter_by(id=story_id).first()
+        if not story:
+            raise HTTPException(status_code=404, detail="Story not found")
+
+        audio_bytes = await stories.export_story_audio(story_id, db)
+        if not audio_bytes:
+            raise HTTPException(status_code=400, detail="Story has no audio items")
+
+        safe_name = "".join(c for c in story.name if c.isalnum() or c in (" ", "-", "_")).strip()
+        if not safe_name:
+            safe_name = "story"
+        filename = f"{safe_name}.wav"
+
+        return StreamingResponse(
+            io.BytesIO(audio_bytes),
+            media_type="audio/wav",
+            headers={"Content-Disposition": safe_content_disposition("attachment", filename)},
+        )
+    except HTTPException:
+        raise
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
@@ -0,0 +1,125 @@
+"""Task and cache management endpoints."""
+
+from datetime import datetime
+
+from fastapi import APIRouter
+
+from .. import models
+from ..utils.cache import clear_voice_prompt_cache
+from ..utils.progress import get_progress_manager
+from ..utils.tasks import get_task_manager
+from fastapi import HTTPException
+
+router = APIRouter()
+
+
+@router.post("/tasks/clear")
+async def clear_all_tasks():
+    """Clear all download tasks and progress state."""
+    task_manager = get_task_manager()
+    progress_manager = get_progress_manager()
+
+    task_manager.clear_all()
+
+    with progress_manager._lock:
+        progress_manager._progress.clear()
+        progress_manager._last_notify_time.clear()
+        progress_manager._last_notify_progress.clear()
+
+    return {"message": "All task state cleared"}
+
+
+@router.post("/cache/clear")
+async def clear_cache():
+    """Clear all voice prompt caches (memory and disk)."""
+    try:
+        deleted_count = clear_voice_prompt_cache()
+        return {
+            "message": "Voice prompt cache cleared successfully",
+            "files_deleted": deleted_count,
+        }
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Failed to clear cache: {str(e)}")
+
+
+@router.get("/tasks/active", response_model=models.ActiveTasksResponse)
+async def get_active_tasks():
+    """Return all currently active downloads and generations."""
+    task_manager = get_task_manager()
+    progress_manager = get_progress_manager()
+
+    active_downloads = []
+    task_manager_downloads = task_manager.get_active_downloads()
+    progress_active = progress_manager.get_all_active()
+
+    download_map = {task.model_name: task for task in task_manager_downloads}
+    progress_map = {p["model_name"]: p for p in progress_active}
+
+    all_model_names = set(download_map.keys()) | set(progress_map.keys())
+    for model_name in all_model_names:
+        task = download_map.get(model_name)
+        progress = progress_map.get(model_name)
+
+        if task:
+            error = task.error
+            if not error:
+                with progress_manager._lock:
+                    pm_data = progress_manager._progress.get(model_name)
+                    if pm_data:
+                        error = pm_data.get("error")
+            prog = progress or {}
+            if not prog:
+                with progress_manager._lock:
+                    pm_data = progress_manager._progress.get(model_name)
+                    if pm_data:
+                        prog = pm_data
+            active_downloads.append(
+                models.ActiveDownloadTask(
+                    model_name=model_name,
+                    status=task.status,
+                    started_at=task.started_at,
+                    error=error,
+                    progress=prog.get("progress"),
+                    current=prog.get("current"),
+                    total=prog.get("total"),
+                    filename=prog.get("filename"),
+                )
+            )
+        elif progress:
+            timestamp_str = progress.get("timestamp")
+            if timestamp_str:
+                try:
+                    started_at = datetime.fromisoformat(timestamp_str.replace("Z", "+00:00"))
+                except (ValueError, AttributeError):
+                    started_at = datetime.utcnow()
+            else:
+                started_at = datetime.utcnow()
+
+            active_downloads.append(
+                models.ActiveDownloadTask(
+                    model_name=model_name,
+                    status=progress.get("status", "downloading"),
+                    started_at=started_at,
+                    error=progress.get("error"),
+                    progress=progress.get("progress"),
+                    current=progress.get("current"),
+                    total=progress.get("total"),
+                    filename=progress.get("filename"),
+                )
+            )
+
+    active_generations = []
+    for gen_task in task_manager.get_active_generations():
+        active_generations.append(
+            models.ActiveGenerationTask(
+                task_id=gen_task.task_id,
+                profile_id=gen_task.profile_id,
+                text_preview=gen_task.text_preview,
+                started_at=gen_task.started_at,
+            )
+        )
+
+    return models.ActiveTasksResponse(
+        downloads=active_downloads,
+        generations=active_generations,
+    )
@@ -0,0 +1,74 @@
+"""Transcription endpoints."""
+
+import asyncio
+import tempfile
+from pathlib import Path
+
+from fastapi import APIRouter, File, Form, HTTPException, UploadFile
+
+from .. import models
+from ..services import transcribe
+from ..services.task_queue import create_background_task
+from ..utils.tasks import get_task_manager
+
+router = APIRouter()
+
+UPLOAD_CHUNK_SIZE = 1024 * 1024  # 1MB
+
+
+@router.post("/transcribe", response_model=models.TranscriptionResponse)
+async def transcribe_audio(
+    file: UploadFile = File(...),
+    language: str | None = Form(None),
+):
+    """Transcribe audio file to text."""
+    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
+        while chunk := await file.read(UPLOAD_CHUNK_SIZE):
+            tmp.write(chunk)
+        tmp_path = tmp.name
+
+    try:
+        from ..utils.audio import load_audio
+
+        audio, sr = await asyncio.to_thread(load_audio, tmp_path)
+        duration = len(audio) / sr
+
+        whisper_model = transcribe.get_whisper_model()
+        model_size = whisper_model.model_size
+
+        if not whisper_model.is_loaded() and not whisper_model._is_model_cached(model_size):
+            progress_model_name = f"whisper-{model_size}"
+            task_manager = get_task_manager()
+
+            async def download_whisper_background():
+                try:
+                    await whisper_model.load_model_async(model_size)
+                    task_manager.complete_download(progress_model_name)
+                except Exception as e:
+                    task_manager.error_download(progress_model_name, str(e))
+
+            task_manager.start_download(progress_model_name)
+            create_background_task(download_whisper_background())
+
+            raise HTTPException(
+                status_code=202,
+                detail={
+                    "message": f"Whisper model {model_size} is being downloaded. Please wait and try again.",
+                    "model_name": progress_model_name,
+                    "downloading": True,
+                },
+            )
+
+        text = await whisper_model.transcribe(tmp_path, language)
+
+        return models.TranscriptionResponse(
+            text=text,
+            duration=duration,
+        )
+
+    except HTTPException:
+        raise
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+    finally:
+        Path(tmp_path).unlink(missing_ok=True)
@@ -6,6 +6,39 @@ absolute imports instead of relative imports.
 """

 import sys
+import os
+
+# On Windows with --noconsole (PyInstaller), sys.stdout/stderr are None.
+# They can also be broken file objects in some edge cases.
+# Redirect to devnull to prevent crashes from print()/tqdm/logging.
+def _is_writable(stream):
+    """Check if a stream is usable for writing."""
+    if stream is None:
+        return False
+    try:
+        stream.write("")
+        return True
+    except Exception:
+        return False
+
+if not _is_writable(sys.stdout):
+    sys.stdout = open(os.devnull, 'w')
+if not _is_writable(sys.stderr):
+    sys.stderr = open(os.devnull, 'w')
+
+# PyInstaller + multiprocessing: child processes re-execute the frozen binary
+# with internal arguments. freeze_support() handles this and exits early.
+import multiprocessing
+multiprocessing.freeze_support()
+
+# In frozen builds, piper_phonemize's espeak-ng C library falls back to
+# /usr/share/espeak-ng-data/ which doesn't exist.  Point it at the bundled
+# data directory instead.
+if getattr(sys, 'frozen', False):
+    _meipass = getattr(sys, '_MEIPASS', os.path.dirname(sys.executable))
+    _espeak_data = os.path.join(_meipass, 'piper_phonemize', 'espeak-ng-data')
+    if os.path.isdir(_espeak_data):
+        os.environ.setdefault('ESPEAK_DATA_PATH', _espeak_data)

 # Fast path: handle --version before any heavy imports so the Rust
 # version check doesn't block for 30+ seconds loading torch etc.
@@ -0,0 +1 @@
+# Services layer — generation orchestration and background task management.
@@ -7,14 +7,14 @@ from datetime import datetime
 import uuid
 from sqlalchemy.orm import Session

-from .models import (
+from ..models import (
    AudioChannelCreate,
    AudioChannelUpdate,
    AudioChannelResponse,
    ChannelVoiceAssignment,
    ProfileChannelAssignment,
 )
-from .database import (
+from ..database import (
    AudioChannel as DBAudioChannel,
    ChannelDeviceMapping as DBChannelDeviceMapping,
    ProfileChannelMapping as DBProfileChannelMapping,
@@ -14,9 +14,9 @@ import sys
 from pathlib import Path
 from typing import Optional

-from .config import get_data_dir
-from .utils.progress import get_progress_manager
-from . import __version__
+from ..config import get_data_dir
+from ..utils.progress import get_progress_manager
+from .. import __version__

 logger = logging.getLogger(__name__)

@@ -199,6 +199,56 @@ async def download_cuda_binary(version: Optional[str] = None):
        raise


+def get_cuda_binary_version() -> Optional[str]:
+    """Get the version of the installed CUDA binary, or None if not installed."""
+    import subprocess
+    cuda_path = get_cuda_binary_path()
+    if not cuda_path:
+        return None
+    try:
+        result = subprocess.run(
+            [str(cuda_path), "--version"],
+            capture_output=True, text=True, timeout=30,
+        )
+        # Output format: "voicebox-server 0.2.0"
+        for line in result.stdout.strip().splitlines():
+            if "voicebox-server" in line:
+                return line.split()[-1]
+    except Exception as e:
+        logger.warning(f"Could not get CUDA binary version: {e}")
+    return None
+
+
+async def check_and_update_cuda_binary():
+    """Check if the CUDA binary is outdated and auto-download if so.
+
+    Called on server startup. If a CUDA binary exists but its version
+    doesn't match the current app version, triggers a background download
+    of the updated CUDA binary. The download progress is visible to the
+    frontend via the existing SSE progress endpoint.
+    """
+    cuda_path = get_cuda_binary_path()
+    if not cuda_path:
+        return  # No CUDA binary installed, nothing to update
+
+    cuda_version = get_cuda_binary_version()
+    current_version = __version__
+
+    if cuda_version == current_version:
+        logger.info(f"CUDA binary is up to date (v{current_version})")
+        return
+
+    logger.info(
+        f"CUDA binary version mismatch: binary=v{cuda_version}, app=v{current_version}. "
+        f"Auto-downloading updated CUDA backend..."
+    )
+
+    try:
+        await download_cuda_binary()
+    except Exception as e:
+        logger.error(f"Auto-update of CUDA binary failed: {e}")
+
+
 async def delete_cuda_binary() -> bool:
    """Delete the downloaded CUDA binary. Returns True if deleted."""
    path = get_cuda_binary_path()
@@ -11,8 +11,8 @@ from typing import List, Optional
 from sqlalchemy.orm import Session
 from sqlalchemy.exc import IntegrityError

-from .database import EffectPreset as DBEffectPreset
-from .models import EffectPresetResponse, EffectPresetCreate, EffectPresetUpdate, EffectConfig
+from ..database import EffectPreset as DBEffectPreset
+from ..models import EffectPresetResponse, EffectPresetCreate, EffectPresetUpdate, EffectConfig


 def _preset_response(p: DBEffectPreset) -> EffectPresetResponse:
@@ -12,16 +12,11 @@ from pathlib import Path
 from typing import Optional
 from sqlalchemy.orm import Session

-from .models import VoiceProfileResponse
-from .database import VoiceProfile as DBVoiceProfile, ProfileSample as DBProfileSample, Generation as DBGeneration, GenerationVersion as DBGenerationVersion
+from ..models import VoiceProfileResponse
+from ..database import VoiceProfile as DBVoiceProfile, ProfileSample as DBProfileSample, Generation as DBGeneration, GenerationVersion as DBGenerationVersion
 from .profiles import create_profile, add_profile_sample
-from .models import VoiceProfileCreate
-from . import config
-
-
-def _get_profiles_dir() -> Path:
-    """Get profiles directory from config."""
-    return config.get_profiles_dir()
+from ..models import VoiceProfileCreate
+from .. import config


 def _get_unique_profile_name(name: str, db: Session) -> str:
@@ -99,7 +94,7 @@ def export_profile_to_zip(profile_id: str, db: Session) -> bytes:

        # Create samples.json mapping
        samples_data = {}
-        profile_dir = _get_profiles_dir() / profile_id
+        profile_dir = config.get_profiles_dir() / profile_id

        for sample in samples:
            # Get filename from audio_path (should be {sample_id}.wav)
@@ -181,7 +176,7 @@ async def import_profile_from_zip(file_bytes: bytes, db: Session) -> VoiceProfil
            profile = await create_profile(profile_create, db)

            # Extract and add samples
-            profile_dir = _get_profiles_dir() / profile.id
+            profile_dir = config.get_profiles_dir() / profile.id
            profile_dir.mkdir(parents=True, exist_ok=True)

            # Handle avatar if present
@@ -351,7 +346,7 @@ async def import_generation_from_zip(file_bytes: bytes, db: Session) -> dict:
    import tempfile
    import shutil
    from datetime import datetime
-    from . import config
+    from .. import config
    
    zip_buffer = io.BytesIO(file_bytes)
    
@@ -0,0 +1,253 @@
+"""
+Unified TTS generation orchestration.
+
+Replaces the three near-identical closures (_run_generation, _run_retry,
+_run_regenerate) that lived in main.py with a single ``run_generation()``
+function parameterized by *mode*.
+
+Mode differences:
+  - "generate"   : full pipeline -- save clean version, optionally apply
+                    effects and create a processed version.
+  - "retry"      : re-runs a failed generation with the same seed.
+                    No effects, no version creation.
+  - "regenerate" : re-runs with seed=None for variation.  Creates a new
+                    version with an auto-incremented "take-N" label.
+"""
+
+from __future__ import annotations
+
+import traceback
+from typing import Literal, Optional
+
+from .. import config
+from . import history, profiles
+from ..database import get_db
+from ..utils.tasks import get_task_manager
+
+
+async def run_generation(
+    *,
+    generation_id: str,
+    profile_id: str,
+    text: str,
+    language: str,
+    engine: str,
+    model_size: str,
+    seed: Optional[int],
+    normalize: bool = False,
+    effects_chain: Optional[list] = None,
+    instruct: Optional[str] = None,
+    mode: Literal["generate", "retry", "regenerate"],
+    max_chunk_chars: Optional[int] = None,
+    crossfade_ms: Optional[int] = None,
+    version_id: Optional[str] = None,
+) -> None:
+    """Execute TTS inference and persist the result.
+
+    This is the single entry point for all background generation work.
+    It is designed to be enqueued via ``services.task_queue.enqueue_generation``.
+    """
+    from ..backends import load_engine_model, get_tts_backend_for_engine, engine_needs_trim
+    from ..utils.chunked_tts import generate_chunked
+    from ..utils.audio import normalize_audio, save_audio, trim_tts_output
+
+    task_manager = get_task_manager()
+    bg_db = next(get_db())
+
+    try:
+        tts_model = get_tts_backend_for_engine(engine)
+
+        if not tts_model.is_loaded():
+            await history.update_generation_status(generation_id, "loading_model", bg_db)
+
+        await load_engine_model(engine, model_size)
+
+        voice_prompt = await profiles.create_voice_prompt_for_profile(
+            profile_id,
+            bg_db,
+            use_cache=True,
+            engine=engine,
+        )
+
+        await history.update_generation_status(generation_id, "generating", bg_db)
+        trim_fn = trim_tts_output if engine_needs_trim(engine) else None
+
+        gen_kwargs: dict = dict(
+            language=language,
+            seed=seed if mode != "regenerate" else None,
+            instruct=instruct,
+            trim_fn=trim_fn,
+        )
+        if max_chunk_chars is not None:
+            gen_kwargs["max_chunk_chars"] = max_chunk_chars
+        if crossfade_ms is not None:
+            gen_kwargs["crossfade_ms"] = crossfade_ms
+
+        audio, sample_rate = await generate_chunked(tts_model, text, voice_prompt, **gen_kwargs)
+
+        # --- Normalize (generate and regenerate always; retry skips) -----
+        if normalize or mode == "regenerate":
+            audio = normalize_audio(audio)
+
+        duration = len(audio) / sample_rate
+
+        # --- Persist audio and update status -----------------------------
+        if mode == "generate":
+            final_path = _save_generate(
+                generation_id=generation_id,
+                audio=audio,
+                sample_rate=sample_rate,
+                effects_chain=effects_chain,
+                save_audio=save_audio,
+                db=bg_db,
+            )
+        elif mode == "retry":
+            final_path = _save_retry(
+                generation_id=generation_id,
+                audio=audio,
+                sample_rate=sample_rate,
+                save_audio=save_audio,
+            )
+        elif mode == "regenerate":
+            final_path = _save_regenerate(
+                generation_id=generation_id,
+                version_id=version_id,
+                audio=audio,
+                sample_rate=sample_rate,
+                save_audio=save_audio,
+                db=bg_db,
+            )
+
+        await history.update_generation_status(
+            generation_id=generation_id,
+            status="completed",
+            db=bg_db,
+            audio_path=final_path,
+            duration=duration,
+        )
+
+    except Exception as e:
+        traceback.print_exc()
+        await history.update_generation_status(
+            generation_id=generation_id,
+            status="failed",
+            db=bg_db,
+            error=str(e),
+        )
+    finally:
+        task_manager.complete_generation(generation_id)
+        bg_db.close()
+
+
+def _save_generate(
+    *,
+    generation_id: str,
+    audio,
+    sample_rate: int,
+    effects_chain: Optional[list],
+    save_audio,
+    db,
+) -> str:
+    """Save clean version and optionally an effects-processed version.
+
+    Returns the final audio path (processed if effects were applied,
+    otherwise clean).
+    """
+    from . import versions as versions_mod
+
+    clean_audio_path = config.get_generations_dir() / f"{generation_id}.wav"
+    save_audio(audio, str(clean_audio_path), sample_rate)
+
+    has_effects = effects_chain and any(e.get("enabled", True) for e in effects_chain)
+
+    versions_mod.create_version(
+        generation_id=generation_id,
+        label="original",
+        audio_path=str(clean_audio_path),
+        db=db,
+        effects_chain=None,
+        is_default=not has_effects,
+    )
+
+    final_audio_path = str(clean_audio_path)
+
+    if has_effects:
+        from ..utils.effects import apply_effects, validate_effects_chain
+
+        error_msg = validate_effects_chain(effects_chain)
+        if error_msg:
+            import logging
+            logging.getLogger(__name__).warning("invalid effects chain, skipping: %s", error_msg)
+            versions_mod.set_default_version(
+                versions_mod.list_versions(generation_id, db)[0].id, db
+            )
+        else:
+            processed_audio = apply_effects(audio, sample_rate, effects_chain)
+            processed_path = config.get_generations_dir() / f"{generation_id}_processed.wav"
+            save_audio(processed_audio, str(processed_path), sample_rate)
+            final_audio_path = str(processed_path)
+            versions_mod.create_version(
+                generation_id=generation_id,
+                label="version-2",
+                audio_path=str(processed_path),
+                db=db,
+                effects_chain=effects_chain,
+                is_default=True,
+            )
+
+    return final_audio_path
+
+
+def _save_retry(
+    *,
+    generation_id: str,
+    audio,
+    sample_rate: int,
+    save_audio,
+) -> str:
+    """Save retry output -- single file, no versions.
+
+    Returns the audio path.
+    """
+    audio_path = config.get_generations_dir() / f"{generation_id}.wav"
+    save_audio(audio, str(audio_path), sample_rate)
+    return str(audio_path)
+
+
+def _save_regenerate(
+    *,
+    generation_id: str,
+    version_id: Optional[str],
+    audio,
+    sample_rate: int,
+    save_audio,
+    db,
+) -> str:
+    """Save regeneration output as a new version with auto-label.
+
+    Returns the audio path.
+    """
+    from . import versions as versions_mod
+
+    import uuid as _uuid
+
+    suffix = _uuid.uuid4().hex[:8]
+    audio_path = config.get_generations_dir() / f"{generation_id}_{suffix}.wav"
+    save_audio(audio, str(audio_path), sample_rate)
+
+    # Count via DB query rather than list length to avoid TOCTOU race
+    from ..database import GenerationVersion as DBGenerationVersion
+
+    count = db.query(DBGenerationVersion).filter_by(generation_id=generation_id).count()
+    label = f"take-{count + 1}"
+
+    versions_mod.create_version(
+        generation_id=generation_id,
+        label=label,
+        audio_path=str(audio_path),
+        db=db,
+        effects_chain=None,
+        is_default=True,
+    )
+
+    return str(audio_path)
@@ -10,14 +10,9 @@ from pathlib import Path
 from sqlalchemy.orm import Session
 from sqlalchemy import or_

-from .models import GenerationRequest, GenerationResponse, HistoryQuery, HistoryResponse, HistoryListResponse, GenerationVersionResponse, EffectConfig
-from .database import Generation as DBGeneration, GenerationVersion as DBGenerationVersion, VoiceProfile as DBVoiceProfile
-from . import config
-
-
-def _get_generations_dir() -> Path:
-    """Get generations directory from config."""
-    return config.get_generations_dir()
+from ..models import GenerationRequest, GenerationResponse, HistoryQuery, HistoryResponse, HistoryListResponse, GenerationVersionResponse, EffectConfig
+from ..database import Generation as DBGeneration, GenerationVersion as DBGenerationVersion, VoiceProfile as DBVoiceProfile
+from .. import config


 def _get_versions_for_generation(generation_id: str, db: Session) -> tuple:
@@ -10,23 +10,23 @@ from pathlib import Path
 from sqlalchemy.orm import Session
 from sqlalchemy import func, select

-from .models import (
+from ..models import (
    VoiceProfileCreate,
    VoiceProfileResponse,
    ProfileSampleCreate,
    ProfileSampleResponse,
 )
-from .database import (
+from ..database import (
    VoiceProfile as DBVoiceProfile,
    ProfileSample as DBProfileSample,
    Generation as DBGeneration,
 )
-from .models import EffectConfig
-from .utils.audio import validate_reference_audio, load_audio, save_audio
-from .utils.images import validate_image, process_avatar
-from .utils.cache import _get_cache_dir, clear_profile_cache
+from ..models import EffectConfig
+from ..utils.audio import validate_reference_audio, load_audio, save_audio
+from ..utils.images import validate_image, process_avatar
+from ..utils.cache import _get_cache_dir, clear_profile_cache
 from .tts import get_tts_model
-from . import config
+from .. import config
 import json as _json


@@ -43,6 +43,7 @@ def _profile_to_response(
            effects_chain = [EffectConfig(**e) for e in raw]
        except Exception as e:
            import logging
+
            logging.warning(f"Failed to parse effects_chain for profile {profile.id}: {e}")
    return VoiceProfileResponse(
        id=profile.id,
@@ -58,11 +59,6 @@ def _profile_to_response(
    )


-def _get_profiles_dir() -> Path:
-    """Get profiles directory from config."""
-    return config.get_profiles_dir()
-
-
 async def create_profile(
    data: VoiceProfileCreate,
    db: Session,
@@ -80,12 +76,10 @@ async def create_profile(
    Raises:
        ValueError: If a profile with the same name already exists
    """
-    # Check if profile name already exists
    existing_profile = db.query(DBVoiceProfile).filter_by(name=data.name).first()
    if existing_profile:
        raise ValueError(f"A profile with the name '{data.name}' already exists. Please choose a different name.")

-    # Create profile in database
    db_profile = DBVoiceProfile(
        id=str(uuid.uuid4()),
        name=data.name,
@@ -99,8 +93,7 @@ async def create_profile(
    db.commit()
    db.refresh(db_profile)

-    # Create profile directory
-    profile_dir = _get_profiles_dir() / db_profile.id
+    profile_dir = config.get_profiles_dir() / db_profile.id
    profile_dir.mkdir(parents=True, exist_ok=True)

    return _profile_to_response(db_profile)
@@ -114,56 +107,50 @@ async def add_profile_sample(
 ) -> ProfileSampleResponse:
    """
    Add a sample to a voice profile.
-    
+
    Args:
        profile_id: Profile ID
        audio_path: Path to temporary audio file
        reference_text: Transcript of audio
        db: Database session
-        
+
    Returns:
        Created sample
    """
-    # Validate profile exists
    profile = db.query(DBVoiceProfile).filter_by(id=profile_id).first()
    if not profile:
        raise ValueError(f"Profile {profile_id} not found")
-    
-    # Validate audio
+
    is_valid, error_msg = validate_reference_audio(audio_path)
    if not is_valid:
        raise ValueError(f"Invalid reference audio: {error_msg}")
-    
-    # Create sample ID and directory
+
    sample_id = str(uuid.uuid4())
-    profile_dir = _get_profiles_dir() / profile_id
+    profile_dir = config.get_profiles_dir() / profile_id
    profile_dir.mkdir(parents=True, exist_ok=True)
-    
-    # Copy audio file to profile directory
+
    dest_path = profile_dir / f"{sample_id}.wav"
    audio, sr = load_audio(audio_path)
    save_audio(audio, str(dest_path), sr)
-    
-    # Create database entry
+
    db_sample = DBProfileSample(
        id=sample_id,
        profile_id=profile_id,
        audio_path=str(dest_path),
        reference_text=reference_text,
    )
-    
+
    db.add(db_sample)
-    
-    # Update profile timestamp
+
    profile.updated_at = datetime.utcnow()
-    
+
    db.commit()
    db.refresh(db_sample)
-    
+
    # Invalidate combined audio cache for this profile
    # Since a new sample was added, any cached combined audio is now stale
    clear_profile_cache(profile_id)
-    
+
    return ProfileSampleResponse.model_validate(db_sample)


@@ -173,18 +160,18 @@ async def get_profile(
 ) -> Optional[VoiceProfileResponse]:
    """
    Get a voice profile by ID.
-    
+
    Args:
        profile_id: Profile ID
        db: Database session
-        
+
    Returns:
        Profile or None if not found
    """
    profile = db.query(DBVoiceProfile).filter_by(id=profile_id).first()
    if not profile:
        return None
-    
+
    return _profile_to_response(profile)


@@ -194,11 +181,11 @@ async def get_profile_samples(
 ) -> List[ProfileSampleResponse]:
    """
    Get all samples for a profile.
-    
+
    Args:
        profile_id: Profile ID
        db: Database session
-        
+
    Returns:
        List of samples
    """
@@ -209,33 +196,27 @@ async def get_profile_samples(
 async def list_profiles(db: Session) -> List[VoiceProfileResponse]:
    """
    List all voice profiles with generation and sample counts.
-    
+
    Args:
        db: Database session
-        
+
    Returns:
        List of profiles
    """
-    profiles = db.query(DBVoiceProfile).order_by(
-        DBVoiceProfile.created_at.desc()
-    ).all()
+    profiles = db.query(DBVoiceProfile).order_by(DBVoiceProfile.created_at.desc()).all()

    if not profiles:
        return []

    # Batch-fetch generation counts
    gen_counts_rows = (
-        db.query(DBGeneration.profile_id, func.count(DBGeneration.id))
-        .group_by(DBGeneration.profile_id)
-        .all()
+        db.query(DBGeneration.profile_id, func.count(DBGeneration.id)).group_by(DBGeneration.profile_id).all()
    )
    gen_counts = {row[0]: row[1] for row in gen_counts_rows}

    # Batch-fetch sample counts
    sample_counts_rows = (
-        db.query(DBProfileSample.profile_id, func.count(DBProfileSample.id))
-        .group_by(DBProfileSample.profile_id)
-        .all()
+        db.query(DBProfileSample.profile_id, func.count(DBProfileSample.id)).group_by(DBProfileSample.profile_id).all()
    )
    sample_counts = {row[0]: row[1] for row in sample_counts_rows}

@@ -272,13 +253,11 @@ async def update_profile(
    if not profile:
        return None

-    # Check if the new name conflicts with another profile
    if profile.name != data.name:
        existing_profile = db.query(DBVoiceProfile).filter_by(name=data.name).first()
        if existing_profile:
            raise ValueError(f"A profile with the name '{data.name}' already exists. Please choose a different name.")

-    # Update fields
    profile.name = data.name
    profile.description = data.description
    profile.language = data.language
@@ -296,33 +275,30 @@ async def delete_profile(
 ) -> bool:
    """
    Delete a voice profile and all associated data.
-    
+
    Args:
        profile_id: Profile ID
        db: Database session
-        
+
    Returns:
        True if deleted, False if not found
    """
    profile = db.query(DBVoiceProfile).filter_by(id=profile_id).first()
    if not profile:
        return False
-    
-    # Delete samples from database
+
    db.query(DBProfileSample).filter_by(profile_id=profile_id).delete()
-    
-    # Delete profile from database
+
    db.delete(profile)
    db.commit()
-    
-    # Delete profile directory
-    profile_dir = _get_profiles_dir() / profile_id
+
+    profile_dir = config.get_profiles_dir() / profile_id
    if profile_dir.exists():
        shutil.rmtree(profile_dir)
-    
+
    # Clean up combined audio cache files for this profile
    clear_profile_cache(profile_id)
-    
+
    return True


@@ -332,34 +308,32 @@ async def delete_profile_sample(
 ) -> bool:
    """
    Delete a profile sample.
-    
+
    Args:
        sample_id: Sample ID
        db: Database session
-        
+
    Returns:
        True if deleted, False if not found
    """
    sample = db.query(DBProfileSample).filter_by(id=sample_id).first()
    if not sample:
        return False
-    
+
    # Store profile_id before deleting
    profile_id = sample.profile_id
-    
-    # Delete audio file
+
    audio_path = Path(sample.audio_path)
    if audio_path.exists():
        audio_path.unlink()
-    
-    # Delete from database
+
    db.delete(sample)
    db.commit()
-    
+
    # Invalidate combined audio cache for this profile
    # Since the sample set changed, any cached combined audio is now stale
    clear_profile_cache(profile_id)
-    
+
    return True


@@ -370,30 +344,30 @@ async def update_profile_sample(
 ) -> Optional[ProfileSampleResponse]:
    """
    Update a profile sample's reference text.
-    
+
    Args:
        sample_id: Sample ID
        reference_text: Updated reference text
        db: Database session
-        
+
    Returns:
        Updated sample or None if not found
    """
    sample = db.query(DBProfileSample).filter_by(id=sample_id).first()
    if not sample:
        return None
-    
+
    # Store profile_id before updating
    profile_id = sample.profile_id
-    
+
    sample.reference_text = reference_text
    db.commit()
    db.refresh(sample)
-    
+
    # Invalidate combined audio cache for this profile
    # Since the reference text changed, cache keys and combined text are now stale
    clear_profile_cache(profile_id)
-    
+
    return ProfileSampleResponse.model_validate(sample)


@@ -415,9 +389,8 @@ async def create_voice_prompt_for_profile(
    Returns:
        Voice prompt dictionary
    """
-    from .backends import get_tts_backend_for_engine
+    from ..backends import get_tts_backend_for_engine

-    # Get all samples for profile
    samples = db.query(DBProfileSample).filter_by(profile_id=profile_id).all()

    if not samples:
@@ -426,7 +399,6 @@ async def create_voice_prompt_for_profile(
    tts_model = get_tts_backend_for_engine(engine)

    if len(samples) == 1:
-        # Single sample - use directly
        sample = samples[0]
        voice_prompt, _ = await tts_model.create_voice_prompt(
            sample.audio_path,
@@ -435,11 +407,9 @@ async def create_voice_prompt_for_profile(
        )
        return voice_prompt
    else:
-        # Multiple samples - combine them
        audio_paths = [s.audio_path for s in samples]
        reference_texts = [s.reference_text for s in samples]

-        # Combine audio
        combined_audio, combined_text = await tts_model.combine_voice_prompts(
            audio_paths,
            reference_texts,
@@ -448,18 +418,16 @@ async def create_voice_prompt_for_profile(
        # Save combined audio to cache directory (persistent)
        # Create a hash of sample IDs to identify this specific combination
        import hashlib
+
        sample_ids_str = "-".join(sorted([s.id for s in samples]))
        combination_hash = hashlib.md5(sample_ids_str.encode()).hexdigest()[:12]
-        
-        # Store in cache directory
+
        cache_dir = _get_cache_dir()
        cache_dir.mkdir(parents=True, exist_ok=True)
        combined_path = cache_dir / f"combined_{profile_id}_{combination_hash}.wav"
-        
-        # Save combined audio
+
        save_audio(combined_audio, str(combined_path), 24000)

-        # Create prompt from combined audio
        voice_prompt, _ = await tts_model.create_voice_prompt(
            str(combined_path),
            combined_text,
@@ -484,17 +452,14 @@ async def upload_avatar(
    Returns:
        Updated profile
    """
-    # Validate profile exists
    profile = db.query(DBVoiceProfile).filter_by(id=profile_id).first()
    if not profile:
        raise ValueError(f"Profile {profile_id} not found")

-    # Validate image
    is_valid, error_msg = validate_image(image_path)
    if not is_valid:
        raise ValueError(error_msg)

-    # Delete existing avatar if present
    if profile.avatar_path:
        old_avatar = Path(profile.avatar_path)
        if old_avatar.exists():
@@ -502,27 +467,22 @@ async def upload_avatar(

    # Determine file extension from uploaded file
    from PIL import Image
+
    with Image.open(image_path) as img:
        # Normalize JPEG variants (MPO is multi-picture format from some cameras)
        img_format = img.format
-        if img_format in ('MPO', 'JPG'):
-            img_format = 'JPEG'
-        
-        ext_map = {
-            'PNG': '.png',
-            'JPEG': '.jpg',
-            'WEBP': '.webp'
-        }
-        ext = ext_map.get(img_format, '.png')
+        if img_format in ("MPO", "JPG"):
+            img_format = "JPEG"

-    # Save processed image to profile directory
-    profile_dir = _get_profiles_dir() / profile_id
+        ext_map = {"PNG": ".png", "JPEG": ".jpg", "WEBP": ".webp"}
+        ext = ext_map.get(img_format, ".png")
+
+    profile_dir = config.get_profiles_dir() / profile_id
    profile_dir.mkdir(parents=True, exist_ok=True)
    output_path = profile_dir / f"avatar{ext}"

    process_avatar(image_path, str(output_path))

-    # Update database
    profile.avatar_path = str(output_path)
    profile.updated_at = datetime.utcnow()

@@ -550,12 +510,10 @@ async def delete_avatar(
    if not profile or not profile.avatar_path:
        return False

-    # Delete avatar file
    avatar_path = Path(profile.avatar_path)
    if avatar_path.exists():
        avatar_path.unlink()

-    # Update database
    profile.avatar_path = None
    profile.updated_at = datetime.utcnow()

@@ -10,7 +10,7 @@ from pathlib import Path
 from sqlalchemy.orm import Session
 from sqlalchemy import func

-from .models import (
+from ..models import (
    StoryCreate,
    StoryResponse,
    StoryDetailResponse,
@@ -22,9 +22,14 @@ from .models import (
    StoryItemSplit,
    StoryItemVersionUpdate,
 )
-from .database import Story as DBStory, StoryItem as DBStoryItem, Generation as DBGeneration, VoiceProfile as DBVoiceProfile
+from ..database import (
+    Story as DBStory,
+    StoryItem as DBStoryItem,
+    Generation as DBGeneration,
+    VoiceProfile as DBVoiceProfile,
+)
 from .history import _get_versions_for_generation
-from .utils.audio import load_audio, save_audio
+from ..utils.audio import load_audio, save_audio
 import numpy as np


@@ -49,11 +54,11 @@ def _build_item_detail(
        id=item.id,
        story_id=item.story_id,
        generation_id=item.generation_id,
-        version_id=getattr(item, 'version_id', None),
+        version_id=getattr(item, "version_id", None),
        start_time_ms=item.start_time_ms,
        track=item.track,
-        trim_start_ms=getattr(item, 'trim_start_ms', 0),
-        trim_end_ms=getattr(item, 'trim_end_ms', 0),
+        trim_start_ms=getattr(item, "trim_start_ms", 0),
+        trim_end_ms=getattr(item, "trim_end_ms", 0),
        created_at=item.created_at,
        profile_id=generation.profile_id,
        profile_name=profile_name,
@@ -95,10 +100,7 @@ async def create_story(
    db.commit()
    db.refresh(db_story)

-    # Get item count
-    item_count = db.query(func.count(DBStoryItem.id)).filter(
-        DBStoryItem.story_id == db_story.id
-    ).scalar()
+    item_count = db.query(func.count(DBStoryItem.id)).filter(DBStoryItem.story_id == db_story.id).scalar()

    response = StoryResponse.model_validate(db_story)
    response.item_count = item_count
@@ -118,17 +120,15 @@ async def list_stories(
        List of stories with item counts
    """
    stories = db.query(DBStory).order_by(DBStory.updated_at.desc()).all()
-    
+
    result = []
    for story in stories:
-        item_count = db.query(func.count(DBStoryItem.id)).filter(
-            DBStoryItem.story_id == story.id
-        ).scalar()
-        
+        item_count = db.query(func.count(DBStoryItem.id)).filter(DBStoryItem.story_id == story.id).scalar()
+
        response = StoryResponse.model_validate(story)
        response.item_count = item_count
        result.append(response)
-    
+
    return result


@@ -150,22 +150,15 @@ async def get_story(
    if not story:
        return None

-    # Get all items ordered by start_time_ms
-    items = db.query(
-        DBStoryItem,
-        DBGeneration,
-        DBVoiceProfile.name.label('profile_name')
-    ).join(
-        DBGeneration,
-        DBStoryItem.generation_id == DBGeneration.id
-    ).join(
-        DBVoiceProfile,
-        DBGeneration.profile_id == DBVoiceProfile.id
-    ).filter(
-        DBStoryItem.story_id == story_id
-    ).order_by(DBStoryItem.start_time_ms).all()
+    items = (
+        db.query(DBStoryItem, DBGeneration, DBVoiceProfile.name.label("profile_name"))
+        .join(DBGeneration, DBStoryItem.generation_id == DBGeneration.id)
+        .join(DBVoiceProfile, DBGeneration.profile_id == DBVoiceProfile.id)
+        .filter(DBStoryItem.story_id == story_id)
+        .order_by(DBStoryItem.start_time_ms)
+        .all()
+    )

-    # Build item details
    item_details = []
    for item, generation, profile_name in items:
        item_details.append(_build_item_detail(item, generation, profile_name, db))
@@ -202,10 +195,7 @@ async def update_story(
    db.commit()
    db.refresh(story)

-    # Get item count
-    item_count = db.query(func.count(DBStoryItem.id)).filter(
-        DBStoryItem.story_id == story.id
-    ).scalar()
+    item_count = db.query(func.count(DBStoryItem.id)).filter(DBStoryItem.story_id == story.id).scalar()

    response = StoryResponse.model_validate(story)
    response.item_count = item_count
@@ -267,10 +257,7 @@ async def add_item_to_story(
        return None

    # Check if generation is already in story
-    existing = db.query(DBStoryItem).filter_by(
-        story_id=story_id,
-        generation_id=data.generation_id
-    ).first()
+    existing = db.query(DBStoryItem).filter_by(story_id=story_id, generation_id=data.generation_id).first()
    if existing:
        # Return existing item
        profile = db.query(DBVoiceProfile).filter_by(id=generation.profile_id).first()
@@ -283,18 +270,16 @@ async def add_item_to_story(
    if data.start_time_ms is not None:
        start_time_ms = data.start_time_ms
    else:
-        # Find the maximum end time on the target track only
-        existing_items = db.query(
-            DBStoryItem,
-            DBGeneration
-        ).join(
-            DBGeneration,
-            DBStoryItem.generation_id == DBGeneration.id
-        ).filter(
-            DBStoryItem.story_id == story_id,
-            DBStoryItem.track == track,
-        ).all()
-        
+        existing_items = (
+            db.query(DBStoryItem, DBGeneration)
+            .join(DBGeneration, DBStoryItem.generation_id == DBGeneration.id)
+            .filter(
+                DBStoryItem.story_id == story_id,
+                DBStoryItem.track == track,
+            )
+            .all()
+        )
+
        if not existing_items:
            start_time_ms = 0
        else:
@@ -302,7 +287,7 @@ async def add_item_to_story(
            for item, gen in existing_items:
                item_end_ms = item.start_time_ms + int(gen.duration * 1000)
                max_end_time_ms = max(max_end_time_ms, item_end_ms)
-            
+
            # Add 200ms gap after the last item
            start_time_ms = max_end_time_ms + 200

@@ -317,10 +302,10 @@ async def add_item_to_story(
    )

    db.add(item)
-    
+
    # Update story updated_at
    story.updated_at = datetime.utcnow()
-    
+
    db.commit()
    db.refresh(item)

@@ -349,10 +334,14 @@ async def move_story_item(
        Updated item detail or None if not found
    """
    # Get the item
-    item = db.query(DBStoryItem).filter_by(
-        id=item_id,
-        story_id=story_id,
-    ).first()
+    item = (
+        db.query(DBStoryItem)
+        .filter_by(
+            id=item_id,
+            story_id=story_id,
+        )
+        .first()
+    )
    if not item:
        return None

@@ -395,10 +384,14 @@ async def remove_item_from_story(
    Returns:
        True if removed, False if not found
    """
-    item = db.query(DBStoryItem).filter_by(
-        id=item_id,
-        story_id=story_id,
-    ).first()
+    item = (
+        db.query(DBStoryItem)
+        .filter_by(
+            id=item_id,
+            story_id=story_id,
+        )
+        .first()
+    )
    if not item:
        return False

@@ -433,10 +426,14 @@ async def trim_story_item(
        Updated item detail or None if not found
    """
    # Get the item
-    item = db.query(DBStoryItem).filter_by(
-        id=item_id,
-        story_id=story_id,
-    ).first()
+    item = (
+        db.query(DBStoryItem)
+        .filter_by(
+            id=item_id,
+            story_id=story_id,
+        )
+        .first()
+    )
    if not item:
        return None

@@ -487,10 +484,14 @@ async def split_story_item(
        List of two updated item details (original and new) or None if not found/invalid
    """
    # Get the item
-    item = db.query(DBStoryItem).filter_by(
-        id=item_id,
-        story_id=story_id,
-    ).first()
+    item = (
+        db.query(DBStoryItem)
+        .filter_by(
+            id=item_id,
+            story_id=story_id,
+        )
+        .first()
+    )
    if not item:
        return None

@@ -500,8 +501,8 @@ async def split_story_item(
        return None

    # Calculate effective duration and validate split point
-    current_trim_start = getattr(item, 'trim_start_ms', 0)
-    current_trim_end = getattr(item, 'trim_end_ms', 0)
+    current_trim_start = getattr(item, "trim_start_ms", 0)
+    current_trim_end = getattr(item, "trim_end_ms", 0)
    original_duration_ms = int(generation.duration * 1000)
    effective_duration_ms = original_duration_ms - current_trim_start - current_trim_end

@@ -520,7 +521,7 @@ async def split_story_item(
        id=str(uuid.uuid4()),
        story_id=story_id,
        generation_id=item.generation_id,  # Same generation, different trim
-        version_id=getattr(item, 'version_id', None),  # Preserve pinned version
+        version_id=getattr(item, "version_id", None),  # Preserve pinned version
        start_time_ms=item.start_time_ms + data.split_time_ms,
        track=item.track,
        trim_start_ms=absolute_split_ms,
@@ -566,10 +567,14 @@ async def duplicate_story_item(
        New item detail or None if not found
    """
    # Get the original item
-    original_item = db.query(DBStoryItem).filter_by(
-        id=item_id,
-        story_id=story_id,
-    ).first()
+    original_item = (
+        db.query(DBStoryItem)
+        .filter_by(
+            id=item_id,
+            story_id=story_id,
+        )
+        .first()
+    )
    if not original_item:
        return None

@@ -579,8 +584,8 @@ async def duplicate_story_item(
        return None

    # Calculate effective duration
-    current_trim_start = getattr(original_item, 'trim_start_ms', 0)
-    current_trim_end = getattr(original_item, 'trim_end_ms', 0)
+    current_trim_start = getattr(original_item, "trim_start_ms", 0)
+    current_trim_end = getattr(original_item, "trim_end_ms", 0)
    original_duration_ms = int(generation.duration * 1000)
    effective_duration_ms = original_duration_ms - current_trim_start - current_trim_end

@@ -589,7 +594,7 @@ async def duplicate_story_item(
        id=str(uuid.uuid4()),
        story_id=story_id,
        generation_id=original_item.generation_id,  # Same generation as original
-        version_id=getattr(original_item, 'version_id', None),  # Preserve pinned version
+        version_id=getattr(original_item, "version_id", None),  # Preserve pinned version
        start_time_ms=original_item.start_time_ms + effective_duration_ms + 200,  # 200ms gap
        track=original_item.track,
        trim_start_ms=current_trim_start,
@@ -673,19 +678,13 @@ async def reorder_story_items(
        return None

    # Get all items for this story with their generation data
-    items_with_gen = db.query(
-        DBStoryItem,
-        DBGeneration,
-        DBVoiceProfile.name.label('profile_name')
-    ).join(
-        DBGeneration,
-        DBStoryItem.generation_id == DBGeneration.id
-    ).join(
-        DBVoiceProfile,
-        DBGeneration.profile_id == DBVoiceProfile.id
-    ).filter(
-        DBStoryItem.story_id == story_id
-    ).all()
+    items_with_gen = (
+        db.query(DBStoryItem, DBGeneration, DBVoiceProfile.name.label("profile_name"))
+        .join(DBGeneration, DBStoryItem.generation_id == DBGeneration.id)
+        .join(DBVoiceProfile, DBGeneration.profile_id == DBVoiceProfile.id)
+        .filter(DBStoryItem.story_id == story_id)
+        .all()
+    )

    # Create maps for quick lookup
    item_map = {item.generation_id: (item, gen, profile_name) for item, gen, profile_name in items_with_gen}
@@ -700,13 +699,13 @@ async def reorder_story_items(

    for gen_id in generation_ids:
        item, generation, profile_name = item_map[gen_id]
-        
+
        # Update the item's start time
        item.start_time_ms = current_time_ms
-        
+
        # Calculate the duration in ms
        duration_ms = int(generation.duration * 1000)
-        
+
        # Move to next position (current end + gap)
        current_time_ms += duration_ms + gap_ms

@@ -738,10 +737,14 @@ async def set_story_item_version(
    Returns:
        Updated item detail or None if not found
    """
-    item = db.query(DBStoryItem).filter_by(
-        id=item_id,
-        story_id=story_id,
-    ).first()
+    item = (
+        db.query(DBStoryItem)
+        .filter_by(
+            id=item_id,
+            story_id=story_id,
+        )
+        .first()
+    )
    if not item:
        return None

@@ -751,11 +754,16 @@ async def set_story_item_version(

    # Validate version_id belongs to this generation if provided
    if data.version_id:
-        from .database import GenerationVersion as DBGenerationVersion
-        version = db.query(DBGenerationVersion).filter_by(
-            id=data.version_id,
-            generation_id=item.generation_id,
-        ).first()
+        from ..database import GenerationVersion as DBGenerationVersion
+
+        version = (
+            db.query(DBGenerationVersion)
+            .filter_by(
+                id=data.version_id,
+                generation_id=item.generation_id,
+            )
+            .first()
+        )
        if not version:
            return None

@@ -793,15 +801,13 @@ async def export_story_audio(
        return None

    # Get all items ordered by start_time_ms
-    items = db.query(
-        DBStoryItem,
-        DBGeneration
-    ).join(
-        DBGeneration,
-        DBStoryItem.generation_id == DBGeneration.id
-    ).filter(
-        DBStoryItem.story_id == story_id
-    ).order_by(DBStoryItem.start_time_ms).all()
+    items = (
+        db.query(DBStoryItem, DBGeneration)
+        .join(DBGeneration, DBStoryItem.generation_id == DBGeneration.id)
+        .filter(DBStoryItem.story_id == story_id)
+        .order_by(DBStoryItem.start_time_ms)
+        .all()
+    )

    if not items:
        return None
@@ -813,8 +819,9 @@ async def export_story_audio(
    for item, generation in items:
        # Resolve audio path: use pinned version if set, otherwise generation default
        resolved_audio_path = generation.audio_path
-        if getattr(item, 'version_id', None):
-            from .database import GenerationVersion as DBGenerationVersion
+        if getattr(item, "version_id", None):
+            from ..database import GenerationVersion as DBGenerationVersion
+
            version = db.query(DBGenerationVersion).filter_by(id=item.version_id).first()
            if version:
                resolved_audio_path = version.audio_path
@@ -826,33 +833,37 @@ async def export_story_audio(
        try:
            audio, sr = load_audio(str(audio_path), sample_rate=sample_rate)
            sample_rate = sr  # Use actual sample rate from first file
-            
+
            # Get trim values
-            trim_start_ms = getattr(item, 'trim_start_ms', 0)
-            trim_end_ms = getattr(item, 'trim_end_ms', 0)
-            
+            trim_start_ms = getattr(item, "trim_start_ms", 0)
+            trim_end_ms = getattr(item, "trim_end_ms", 0)
+
            # Calculate effective duration
            original_duration_ms = int(generation.duration * 1000)
            effective_duration_ms = original_duration_ms - trim_start_ms - trim_end_ms
-            
+
            # Slice audio based on trim values
            trim_start_sample = int((trim_start_ms / 1000.0) * sample_rate)
            trim_end_sample = int((trim_end_ms / 1000.0) * sample_rate)
-            
+
            # Extract the trimmed portion
            if trim_end_ms > 0:
-                trimmed_audio = audio[trim_start_sample:-trim_end_sample] if trim_end_sample > 0 else audio[trim_start_sample:]
+                trimmed_audio = (
+                    audio[trim_start_sample:-trim_end_sample] if trim_end_sample > 0 else audio[trim_start_sample:]
+                )
            else:
                trimmed_audio = audio[trim_start_sample:]
-            
+
            # Store audio with its timecode info
            start_time_ms = item.start_time_ms
-            
-            audio_data.append({
-                'audio': trimmed_audio,
-                'start_time_ms': start_time_ms,
-                'duration_ms': effective_duration_ms,
-            })
+
+            audio_data.append(
+                {
+                    "audio": trimmed_audio,
+                    "start_time_ms": start_time_ms,
+                    "duration_ms": effective_duration_ms,
+                }
+            )
        except Exception:
            # Skip files that can't be loaded
            continue
@@ -861,33 +872,30 @@ async def export_story_audio(
        return None

    # Calculate total duration: max(start_time_ms + duration_ms)
-    max_end_time_ms = max(
-        (data['start_time_ms'] + data['duration_ms'] for data in audio_data),
-        default=0
-    )
-    
+    max_end_time_ms = max((data["start_time_ms"] + data["duration_ms"] for data in audio_data), default=0)
+
    # Convert to samples
    total_samples = int((max_end_time_ms / 1000.0) * sample_rate)
-    
+
    # Create output buffer initialized to zeros
    final_audio = np.zeros(total_samples, dtype=np.float32)

    # Mix each audio segment at its timecode position
    for data in audio_data:
-        audio = data['audio']
-        start_time_ms = data['start_time_ms']
-        
+        audio = data["audio"]
+        start_time_ms = data["start_time_ms"]
+
        # Calculate start sample index
        start_sample = int((start_time_ms / 1000.0) * sample_rate)
-        
+
        # Ensure we don't exceed buffer bounds
        audio_length = len(audio)
        end_sample = min(start_sample + audio_length, total_samples)
-        
+
        if start_sample < total_samples:
            # Trim audio if it extends beyond buffer
-            audio_to_mix = audio[:end_sample - start_sample]
-            
+            audio_to_mix = audio[: end_sample - start_sample]
+
            # Mix: add audio to existing buffer (overlapping audio will sum)
            # Normalize to prevent clipping (simple approach: divide by max)
            final_audio[start_sample:end_sample] += audio_to_mix
@@ -898,14 +906,14 @@ async def export_story_audio(
        final_audio = final_audio / max_val

    # Save to temporary file
-    with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as tmp:
+    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
        tmp_path = tmp.name

    try:
        save_audio(final_audio, tmp_path, sample_rate)

        # Read file bytes
-        with open(tmp_path, 'rb') as f:
+        with open(tmp_path, "rb") as f:
            audio_bytes = f.read()

        return audio_bytes
@@ -0,0 +1,48 @@
+"""
+Serial generation queue — ensures only one TTS inference runs at a time
+to avoid GPU contention.
+"""
+
+import asyncio
+import traceback
+
+# Keep references to fire-and-forget background tasks to prevent GC
+_background_tasks: set = set()
+
+# Generation queue — serializes TTS inference to avoid GPU contention
+_generation_queue: asyncio.Queue = None  # type: ignore  # initialized at startup
+
+
+def create_background_task(coro) -> asyncio.Task:
+    """Create a background task and prevent it from being garbage collected."""
+    task = asyncio.create_task(coro)
+    _background_tasks.add(task)
+    task.add_done_callback(_background_tasks.discard)
+    return task
+
+
+async def _generation_worker():
+    """Worker that processes generation tasks one at a time."""
+    while True:
+        coro = await _generation_queue.get()
+        try:
+            await coro
+        except Exception:
+            traceback.print_exc()
+        finally:
+            _generation_queue.task_done()
+
+
+def enqueue_generation(coro):
+    """Add a generation coroutine to the serial queue."""
+    _generation_queue.put_nowait(coro)
+
+
+def init_queue():
+    """Initialize the generation queue and start the worker.
+
+    Must be called once during application startup (inside a running event loop).
+    """
+    global _generation_queue
+    _generation_queue = asyncio.Queue()
+    create_background_task(_generation_worker())
@@ -3,7 +3,7 @@ STT (Speech-to-Text) module - delegates to backend abstraction layer.
 """

 from typing import Optional
-from .backends import get_stt_backend, STTBackend
+from ..backends import get_stt_backend, STTBackend


 def get_whisper_model() -> STTBackend:
@@ -7,7 +7,7 @@ import numpy as np
 import io
 import soundfile as sf

-from .backends import get_tts_backend, TTSBackend
+from ..backends import get_tts_backend, TTSBackend


 def get_tts_model() -> TTSBackend:
@@ -14,12 +14,12 @@ from typing import List, Optional

 from sqlalchemy.orm import Session

-from .database import (
+from ..database import (
    GenerationVersion as DBGenerationVersion,
    Generation as DBGeneration,
 )
-from .models import GenerationVersionResponse, EffectConfig
-from . import config
+from ..models import GenerationVersionResponse, EffectConfig
+from .. import config


 def _version_response(v: DBGenerationVersion) -> GenerationVersionResponse:
@@ -1,66 +0,0 @@
-"""
-Audio studio module for timeline editing.
-"""
-
-from typing import List, Dict, Optional
-import numpy as np
-
-
-class AudioStudio:
-    """Audio editing and timeline management."""
-    
-    async def get_word_timestamps(
-        self,
-        audio_path: str,
-        text: str,
-    ) -> List[Dict[str, float]]:
-        """
-        Get word-level timestamps for audio.
-        
-        Args:
-            audio_path: Path to audio file
-            text: Corresponding text
-            
-        Returns:
-            List of word timestamps: [{"word": "...", "start": 0.0, "end": 0.5}, ...]
-        """
-        # TODO: Implement Whisper alignment
-        raise NotImplementedError("Word timestamps not yet implemented")
-    
-    async def mix_audio(
-        self,
-        audio_paths: List[str],
-        volumes: Optional[List[float]] = None,
-    ) -> bytes:
-        """
-        Mix multiple audio files together.
-        
-        Args:
-            audio_paths: List of audio file paths
-            volumes: Optional volume levels (0.0-1.0) for each track
-            
-        Returns:
-            Mixed audio bytes (WAV format)
-        """
-        # TODO: Implement audio mixing
-        raise NotImplementedError("Audio mixing not yet implemented")
-    
-    async def trim_audio(
-        self,
-        audio_path: str,
-        start: float,
-        end: float,
-    ) -> bytes:
-        """
-        Trim audio to specified time range.
-        
-        Args:
-            audio_path: Path to audio file
-            start: Start time in seconds
-            end: End time in seconds
-            
-        Returns:
-            Trimmed audio bytes (WAV format)
-        """
-        # TODO: Implement audio trimming
-        raise NotImplementedError("Audio trimming not yet implemented")
@@ -37,11 +37,10 @@ async def monitor_sse_stream(model_name: str, timeout: int = 120):
                    if line.startswith("data: "):
                        try:
                            data = json.loads(line[6:])
-                            print(f"[{timestamp}] → SSE Event: {data['status']:12} {data.get('progress', 0):6.1f}% {data.get('filename', '')}")
-                            events.append({
-                                **data,
-                                "_timestamp": timestamp
-                            })
+                            print(
+                                f"[{timestamp}] → SSE Event: {data['status']:12} {data.get('progress', 0):6.1f}% {data.get('filename', '')}"
+                            )
+                            events.append({**data, "_timestamp": timestamp})

                            # Stop if complete or error
                            if data.get("status") in ("complete", "error"):
@@ -74,12 +73,15 @@ async def trigger_generation(profile_id: str, text: str, model_size: str = "1.7B

    try:
        async with httpx.AsyncClient(timeout=120) as client:
-            response = await client.post(url, json={
-                "profile_id": profile_id,
-                "text": text,
-                "language": "en",
-                "model_size": model_size,
-            })
+            response = await client.post(
+                url,
+                json={
+                    "profile_id": profile_id,
+                    "text": text,
+                    "language": "en",
+                    "model_size": model_size,
+                },
+            )

            print(f"[{_timestamp()}] Response: {response.status_code}")

@@ -140,7 +142,7 @@ def _timestamp():
 async def test_generation_with_cached_model():
    """
    Test Case 1: Generation when model is already cached.
-    
+
    This should NOT show any download progress events.
    If it does, that's the UX bug we're trying to fix.
    """
@@ -194,7 +196,7 @@ async def test_generation_with_cached_model():
 async def test_generation_with_fresh_download():
    """
    Test Case 2: Generation when model needs to be downloaded.
-    
+
    This SHOULD show download progress events.
    """
    print("\n" + "=" * 80)
@@ -292,24 +294,6 @@ async def main():
        print("  Users see progress events even when the model is already cached,")
        print("  making them think the model is downloading again.")

-    # Test Case 2: Fresh download (optional, commented out by default)
-    # Uncomment if you want to test download progress
-    # print("\n" + "🧪 " * 20)
-    # events_download = await test_generation_with_fresh_download()
-    #
-    # print("\n" + "=" * 80)
-    # print("TEST CASE 2 RESULTS: Generation with Model Download")
-    # print("=" * 80)
-    #
-    # if not events_download:
-    #     print("ℹ Model was already cached, no download occurred")
-    # else:
-    #     print(f"✓ Received {len(events_download)} download progress events")
-    #     print("\nDownload Timeline:")
-    #     for i, event in enumerate(events_download, 1):
-    #         timestamp = event.pop("_timestamp", "??:??:??.???")
-    #         print(f"  {i}. [{timestamp}] {event}")
-
    print("\n" + "=" * 80)
    print("Test Complete!")
    print("=" * 80)
@@ -45,8 +45,8 @@ def test_db():
@pytest.fixture
 def mock_profiles_dir(monkeypatch, tmp_path):
    """Mock the profiles directory to use a temporary path."""
-    import profiles
-    monkeypatch.setattr(profiles, '_get_profiles_dir', lambda: tmp_path)
+    from backend import config
+    monkeypatch.setattr(config, 'get_profiles_dir', lambda: tmp_path)
    return tmp_path


@@ -3,12 +3,15 @@ Voice prompt caching utilities.
 """

 import hashlib
+import logging
 import torch
 from pathlib import Path
 from typing import Optional, Union, Dict, Any

 from .. import config

+logger = logging.getLogger(__name__)
+

 def _get_cache_dir() -> Path:
    """Get cache directory from config."""
@@ -93,17 +96,17 @@ def cache_voice_prompt(
 def clear_voice_prompt_cache() -> int:
    """
    Clear all voice prompt caches (memory and disk).
-    
+
    Returns:
        Number of cache files deleted
    """
    # Clear memory cache
    _memory_cache.clear()
-    
+
    # Clear disk cache
    cache_dir = _get_cache_dir()
    deleted_count = 0
-    
+
    if cache_dir.exists():
        # Delete prompt cache files
        for cache_file in cache_dir.glob("*.prompt"):
@@ -111,32 +114,32 @@ def clear_voice_prompt_cache() -> int:
                cache_file.unlink()
                deleted_count += 1
            except Exception as e:
-                print(f"Failed to delete cache file {cache_file}: {e}")
-        
+                logger.warning("Failed to delete cache file %s: %s", cache_file, e)
+
        # Delete combined audio files
        for audio_file in cache_dir.glob("combined_*.wav"):
            try:
                audio_file.unlink()
                deleted_count += 1
            except Exception as e:
-                print(f"Failed to delete combined audio file {audio_file}: {e}")
-    
+                logger.warning("Failed to delete combined audio file %s: %s", audio_file, e)
+
    return deleted_count


 def clear_profile_cache(profile_id: str) -> int:
    """
    Clear cache files for a specific profile.
-    
+
    Args:
        profile_id: Profile ID
-    
+
    Returns:
        Number of cache files deleted
    """
    cache_dir = _get_cache_dir()
    deleted_count = 0
-    
+
    if cache_dir.exists():
        # Delete combined audio files for this profile
        pattern = f"combined_{profile_id}_*.wav"
@@ -145,6 +148,6 @@ def clear_profile_cache(profile_id: str) -> int:
                audio_file.unlink()
                deleted_count += 1
            except Exception as e:
-                print(f"Failed to delete combined audio file {audio_file}: {e}")
-    
+                logger.warning("Failed to delete combined audio file %s: %s", audio_file, e)
+
    return deleted_count
@@ -58,11 +58,6 @@ _ABBREVIATIONS = frozenset(
 _PARA_TAG_RE = re.compile(r"\[[^\]]*\]")


-# ---------------------------------------------------------------------------
-# Text splitting
-# ---------------------------------------------------------------------------
-
-
 def split_text_into_chunks(text: str, max_chars: int = DEFAULT_MAX_CHUNK_CHARS) -> List[str]:
    """Split *text* at natural boundaries into chunks of at most *max_chars*.

@@ -174,11 +169,6 @@ def _safe_hard_cut(segment: str, max_chars: int) -> int:
    return cut


-# ---------------------------------------------------------------------------
-# Audio concatenation
-# ---------------------------------------------------------------------------
-
-
 def concatenate_audio_chunks(
    chunks: List[np.ndarray],
    sample_rate: int,
@@ -211,11 +201,6 @@ def concatenate_audio_chunks(
    return result


-# ---------------------------------------------------------------------------
-# Engine-agnostic chunked generation
-# ---------------------------------------------------------------------------
-
-
 async def generate_chunked(
    backend,
    text: str,
@@ -264,7 +249,11 @@ async def generate_chunked(
    if len(chunks) <= 1:
        # Short text — single-shot fast path
        audio, sample_rate = await backend.generate(
-            text, voice_prompt, language, seed, instruct,
+            text,
+            voice_prompt,
+            language,
+            seed,
+            instruct,
        )
        if trim_fn is not None:
            audio = trim_fn(audio, sample_rate)
@@ -273,7 +262,9 @@ async def generate_chunked(
    # Long text — chunked generation
    logger.info(
        "Splitting %d chars into %d chunks (max %d chars each)",
-        len(text), len(chunks), max_chunk_chars,
+        len(text),
+        len(chunks),
+        max_chunk_chars,
    )
    audio_chunks: List[np.ndarray] = []
    sample_rate: int | None = None
@@ -281,7 +272,9 @@ async def generate_chunked(
    for i, chunk_text in enumerate(chunks):
        logger.info(
            "Generating chunk %d/%d (%d chars)",
-            i + 1, len(chunks), len(chunk_text),
+            i + 1,
+            len(chunks),
+            len(chunk_text),
        )
        # Vary the seed per chunk to avoid correlated RNG artefacts,
        # but keep it deterministic so the same (text, seed) pair
@@ -289,7 +282,11 @@ async def generate_chunked(
        chunk_seed = (seed + i) if seed is not None else None

        chunk_audio, chunk_sr = await backend.generate(
-            chunk_text, voice_prompt, language, chunk_seed, instruct,
+            chunk_text,
+            voice_prompt,
+            language,
+            chunk_seed,
+            instruct,
        )
        if trim_fn is not None:
            chunk_audio = trim_fn(chunk_audio, chunk_sr)
@@ -35,10 +35,6 @@ from pedalboard import (
 )


-# ---------------------------------------------------------------------------
-# Effect registry: maps type names -> (pedalboard class, param definitions)
-# ---------------------------------------------------------------------------
-
 # Each param definition: (default, min, max, description)
 EFFECT_REGISTRY: Dict[str, Dict[str, Any]] = {
    "chorus": {
@@ -46,11 +42,17 @@ EFFECT_REGISTRY: Dict[str, Dict[str, Any]] = {
        "label": "Chorus / Flanger",
        "description": "Modulated delay for flanging or chorus effects. Short centre_delay_ms (<10) gives flanger; longer gives chorus.",
        "params": {
-            "rate_hz":          {"default": 1.0,  "min": 0.01, "max": 20.0,  "step": 0.01, "description": "LFO speed (Hz)"},
-            "depth":            {"default": 0.5,  "min": 0.0,  "max": 1.0,   "step": 0.01, "description": "Modulation depth"},
-            "feedback":         {"default": 0.0,  "min": 0.0,  "max": 0.95,  "step": 0.01, "description": "Feedback amount"},
-            "centre_delay_ms":  {"default": 7.0,  "min": 0.5,  "max": 50.0,  "step": 0.1,  "description": "Centre delay (ms)"},
-            "mix":              {"default": 0.5,  "min": 0.0,  "max": 1.0,   "step": 0.01, "description": "Wet/dry mix"},
+            "rate_hz": {"default": 1.0, "min": 0.01, "max": 20.0, "step": 0.01, "description": "LFO speed (Hz)"},
+            "depth": {"default": 0.5, "min": 0.0, "max": 1.0, "step": 0.01, "description": "Modulation depth"},
+            "feedback": {"default": 0.0, "min": 0.0, "max": 0.95, "step": 0.01, "description": "Feedback amount"},
+            "centre_delay_ms": {
+                "default": 7.0,
+                "min": 0.5,
+                "max": 50.0,
+                "step": 0.1,
+                "description": "Centre delay (ms)",
+            },
+            "mix": {"default": 0.5, "min": 0.0, "max": 1.0, "step": 0.01, "description": "Wet/dry mix"},
        },
    },
    "reverb": {
@@ -58,11 +60,11 @@ EFFECT_REGISTRY: Dict[str, Dict[str, Any]] = {
        "label": "Reverb",
        "description": "Room reverb effect.",
        "params": {
-            "room_size":  {"default": 0.5,  "min": 0.0,  "max": 1.0,  "step": 0.01, "description": "Room size"},
-            "damping":    {"default": 0.5,  "min": 0.0,  "max": 1.0,  "step": 0.01, "description": "High frequency damping"},
-            "wet_level":  {"default": 0.33, "min": 0.0,  "max": 1.0,  "step": 0.01, "description": "Wet level"},
-            "dry_level":  {"default": 0.4,  "min": 0.0,  "max": 1.0,  "step": 0.01, "description": "Dry level"},
-            "width":      {"default": 1.0,  "min": 0.0,  "max": 1.0,  "step": 0.01, "description": "Stereo width"},
+            "room_size": {"default": 0.5, "min": 0.0, "max": 1.0, "step": 0.01, "description": "Room size"},
+            "damping": {"default": 0.5, "min": 0.0, "max": 1.0, "step": 0.01, "description": "High frequency damping"},
+            "wet_level": {"default": 0.33, "min": 0.0, "max": 1.0, "step": 0.01, "description": "Wet level"},
+            "dry_level": {"default": 0.4, "min": 0.0, "max": 1.0, "step": 0.01, "description": "Dry level"},
+            "width": {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01, "description": "Stereo width"},
        },
    },
    "delay": {
@@ -70,9 +72,15 @@ EFFECT_REGISTRY: Dict[str, Dict[str, Any]] = {
        "label": "Delay",
        "description": "Echo / delay line.",
        "params": {
-            "delay_seconds": {"default": 0.3,  "min": 0.01, "max": 2.0,  "step": 0.01, "description": "Delay time (seconds)"},
-            "feedback":      {"default": 0.3,  "min": 0.0,  "max": 0.95, "step": 0.01, "description": "Feedback amount"},
-            "mix":           {"default": 0.3,  "min": 0.0,  "max": 1.0,  "step": 0.01, "description": "Wet/dry mix"},
+            "delay_seconds": {
+                "default": 0.3,
+                "min": 0.01,
+                "max": 2.0,
+                "step": 0.01,
+                "description": "Delay time (seconds)",
+            },
+            "feedback": {"default": 0.3, "min": 0.0, "max": 0.95, "step": 0.01, "description": "Feedback amount"},
+            "mix": {"default": 0.3, "min": 0.0, "max": 1.0, "step": 0.01, "description": "Wet/dry mix"},
        },
    },
    "compressor": {
@@ -80,10 +88,16 @@ EFFECT_REGISTRY: Dict[str, Dict[str, Any]] = {
        "label": "Compressor",
        "description": "Dynamic range compression for consistent loudness.",
        "params": {
-            "threshold_db": {"default": -20.0, "min": -60.0, "max": 0.0,   "step": 0.5, "description": "Threshold (dB)"},
-            "ratio":        {"default": 4.0,   "min": 1.0,   "max": 20.0,  "step": 0.1, "description": "Compression ratio"},
-            "attack_ms":    {"default": 10.0,  "min": 0.1,   "max": 100.0, "step": 0.1, "description": "Attack time (ms)"},
-            "release_ms":   {"default": 100.0, "min": 10.0,  "max": 1000.0,"step": 1.0, "description": "Release time (ms)"},
+            "threshold_db": {"default": -20.0, "min": -60.0, "max": 0.0, "step": 0.5, "description": "Threshold (dB)"},
+            "ratio": {"default": 4.0, "min": 1.0, "max": 20.0, "step": 0.1, "description": "Compression ratio"},
+            "attack_ms": {"default": 10.0, "min": 0.1, "max": 100.0, "step": 0.1, "description": "Attack time (ms)"},
+            "release_ms": {
+                "default": 100.0,
+                "min": 10.0,
+                "max": 1000.0,
+                "step": 1.0,
+                "description": "Release time (ms)",
+            },
        },
    },
    "gain": {
@@ -99,7 +113,13 @@ EFFECT_REGISTRY: Dict[str, Dict[str, Any]] = {
        "label": "High-Pass Filter",
        "description": "Removes frequencies below the cutoff.",
        "params": {
-            "cutoff_frequency_hz": {"default": 80.0, "min": 20.0, "max": 8000.0, "step": 1.0, "description": "Cutoff frequency (Hz)"},
+            "cutoff_frequency_hz": {
+                "default": 80.0,
+                "min": 20.0,
+                "max": 8000.0,
+                "step": 1.0,
+                "description": "Cutoff frequency (Hz)",
+            },
        },
    },
    "lowpass": {
@@ -107,7 +127,13 @@ EFFECT_REGISTRY: Dict[str, Dict[str, Any]] = {
        "label": "Low-Pass Filter",
        "description": "Removes frequencies above the cutoff.",
        "params": {
-            "cutoff_frequency_hz": {"default": 8000.0, "min": 200.0, "max": 20000.0, "step": 1.0, "description": "Cutoff frequency (Hz)"},
+            "cutoff_frequency_hz": {
+                "default": 8000.0,
+                "min": 200.0,
+                "max": 20000.0,
+                "step": 1.0,
+                "description": "Cutoff frequency (Hz)",
+            },
        },
    },
    "pitch_shift": {
@@ -121,10 +147,6 @@ EFFECT_REGISTRY: Dict[str, Dict[str, Any]] = {
 }


-# ---------------------------------------------------------------------------
-# Built-in presets
-# ---------------------------------------------------------------------------
-
 BUILTIN_PRESETS: Dict[str, Dict[str, Any]] = {
    "robotic": {
        "name": "Robotic",
@@ -233,10 +255,6 @@ BUILTIN_PRESETS: Dict[str, Dict[str, Any]] = {
 }


-# ---------------------------------------------------------------------------
-# Public API
-# ---------------------------------------------------------------------------
-
 def get_available_effects() -> List[Dict[str, Any]]:
    """Return the list of available effect types with their parameter definitions.

@@ -244,15 +262,14 @@ def get_available_effects() -> List[Dict[str, Any]]:
    """
    result = []
    for effect_type, info in EFFECT_REGISTRY.items():
-        result.append({
-            "type": effect_type,
-            "label": info["label"],
-            "description": info["description"],
-            "params": {
-                name: {k: v for k, v in pdef.items()}
-                for name, pdef in info["params"].items()
-            },
-        })
+        result.append(
+            {
+                "type": effect_type,
+                "label": info["label"],
+                "description": info["description"],
+                "params": {name: {k: v for k, v in pdef.items()} for name, pdef in info["params"].items()},
+            }
+        )
    return result


@@ -1,26 +1,26 @@
-"""
-Monkey patch for huggingface_hub to force offline mode with cached models.
-This prevents mlx_audio from making network requests when models are already downloaded.
+"""Monkey-patch huggingface_hub to force offline mode with cached models.
+
+Prevents mlx_audio from making network requests when models are already
+downloaded. Must be imported BEFORE mlx_audio.
 """

+import logging
 import os
 from pathlib import Path
 from typing import Optional, Union

+logger = logging.getLogger(__name__)
+

 def patch_huggingface_hub_offline():
-    """
-    Monkey-patch huggingface_hub to force offline mode.
-    This must be called BEFORE importing mlx_audio.
-    """
+    """Monkey-patch huggingface_hub to force offline mode."""
    try:
-        import huggingface_hub
+        import huggingface_hub  # noqa: F401 -- need the package loaded
        from huggingface_hub import constants as hf_constants
        from huggingface_hub.file_download import _try_to_load_from_cache
-        
-        # Store original function
+
        original_try_load = _try_to_load_from_cache
-        
+
        def _patched_try_to_load_from_cache(
            repo_id: str,
            filename: str,
@@ -28,11 +28,6 @@ def patch_huggingface_hub_offline():
            revision: Optional[str] = None,
            repo_type: Optional[str] = None,
        ):
-            """
-            Patched version that forces offline mode.
-            Returns None if not cached (instead of making network request).
-            """
-            # Always use the original function, but we're already in HF_HUB_OFFLINE mode
            result = original_try_load(
                repo_id=repo_id,
                filename=filename,
@@ -40,61 +35,54 @@ def patch_huggingface_hub_offline():
                revision=revision,
                repo_type=repo_type,
            )
-            
+
            if result is None:
-                # File not in cache - log this for debugging
                cache_path = Path(hf_constants.HF_HUB_CACHE) / f"models--{repo_id.replace('/', '--')}"
-                print(f"[HF_PATCH] File not cached: {repo_id}/{filename}")
-                print(f"[HF_PATCH] Expected at: {cache_path}")
+                logger.debug("file not cached: %s/%s (expected at %s)", repo_id, filename, cache_path)
            else:
-                print(f"[HF_PATCH] Cache hit: {repo_id}/{filename}")
-            
+                logger.debug("cache hit: %s/%s", repo_id, filename)
+
            return result
-        
-        # Replace the function
+
        import huggingface_hub.file_download as fd
+
        fd._try_to_load_from_cache = _patched_try_to_load_from_cache
-        
-        print("[HF_PATCH] huggingface_hub patched for offline mode")
-        
+        logger.debug("huggingface_hub patched for offline mode")
+
    except ImportError:
-        print("[HF_PATCH] huggingface_hub not found, skipping patch")
-    except Exception as e:
-        print(f"[HF_PATCH] Error patching huggingface_hub: {e}")
+        logger.debug("huggingface_hub not available, skipping offline patch")
+    except Exception:
+        logger.exception("failed to patch huggingface_hub for offline mode")


 def ensure_original_qwen_config_cached():
+    """Symlink the original Qwen repo cache to the MLX community version.
+
+    mlx_audio may try to fetch config from the original Qwen repo. If only
+    the MLX community variant is cached, create a symlink so the cache lookup
+    succeeds without a network request.
    """
-    The MLX community model is based on the original Qwen model.
-    mlx_audio may try to fetch config from the original repo.
-    We need to ensure that config is available in the cache.
-    """
-    from huggingface_hub import constants as hf_constants
-    
-    # Original Qwen model that mlx_audio might reference
+    try:
+        from huggingface_hub import constants as hf_constants
+    except ImportError:
+        return
+
    original_repo = "Qwen/Qwen3-TTS-12Hz-1.7B-Base"
    mlx_repo = "mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16"
-    
+
    cache_dir = Path(hf_constants.HF_HUB_CACHE)
-    
    original_path = cache_dir / f"models--{original_repo.replace('/', '--')}"
    mlx_path = cache_dir / f"models--{mlx_repo.replace('/', '--')}"
-    
-    # If original repo cache doesn't exist but MLX does, create a symlink or copy config
+
    if not original_path.exists() and mlx_path.exists():
-        print(f"[HF_PATCH] Original repo not cached, but MLX version is")
-        print(f"[HF_PATCH] Creating symlink from {original_repo} -> {mlx_repo}")
-        
        try:
-            # Create a symlink so the cache lookup succeeds
            original_path.parent.mkdir(parents=True, exist_ok=True)
            original_path.symlink_to(mlx_path, target_is_directory=True)
-            print(f"[HF_PATCH] Symlink created successfully")
-        except Exception as e:
-            print(f"[HF_PATCH] Could not create symlink: {e}")
+            logger.info("created cache symlink: %s -> %s", original_repo, mlx_repo)
+        except Exception:
+            logger.warning("could not create cache symlink for %s", original_repo, exc_info=True)


-# Auto-apply patch when module is imported
 if os.environ.get("VOICEBOX_OFFLINE_PATCH", "1") != "0":
    patch_huggingface_hub_offline()
    ensure_original_qwen_config_cached()
@@ -4,13 +4,16 @@ HuggingFace Hub download progress tracking.

 from typing import Optional, Callable
 from contextlib import contextmanager
+import logging
 import threading
 import sys

+logger = logging.getLogger(__name__)
+

 class HFProgressTracker:
    """Tracks HuggingFace Hub download progress by intercepting tqdm."""
-    
+
    def __init__(self, progress_callback: Optional[Callable] = None, filter_non_downloads: bool = False):
        self.progress_callback = progress_callback
        self.filter_non_downloads = filter_non_downloads  # Only filter if True
@@ -23,12 +26,12 @@ class HFProgressTracker:
        self._current_filename = ""
        self._active_tqdms = {}  # Track active tqdm instances
        self._hf_tqdm_original_update = None  # For monkey-patching hf's tqdm
-    
+
    def _create_tracked_tqdm_class(self):
        """Create a tqdm subclass that tracks progress."""
        tracker = self
        original_tqdm = self._original_tqdm_class
-        
+
        class TrackedTqdm(original_tqdm):
            """A tqdm subclass that reports progress to our tracker."""

@@ -39,7 +42,7 @@ class HFProgressTracker:
                    first_arg = args[0]
                    if isinstance(first_arg, str):
                        desc = first_arg
-                
+
                filename = ""
                if desc:
                    # Try to extract filename from description
@@ -48,38 +51,68 @@ class HFProgressTracker:
                        filename = desc.split(":")[0].strip()
                    else:
                        filename = desc.strip()
-                
+
                # Filter out non-standard kwargs that huggingface_hub might pass
                # These are custom kwargs that tqdm doesn't understand
                filtered_kwargs = {}
                # Known tqdm kwargs - pass these through
                tqdm_kwargs = {
-                    'iterable', 'desc', 'total', 'leave', 'file', 'ncols', 'mininterval',
-                    'maxinterval', 'miniters', 'ascii', 'disable', 'unit', 'unit_scale',
-                    'dynamic_ncols', 'smoothing', 'bar_format', 'initial', 'position',
-                    'postfix', 'unit_divisor', 'write_bytes', 'lock_args', 'nrows',
-                    'colour', 'color', 'delay', 'gui', 'disable_default', 'pos'
+                    "iterable",
+                    "desc",
+                    "total",
+                    "leave",
+                    "file",
+                    "ncols",
+                    "mininterval",
+                    "maxinterval",
+                    "miniters",
+                    "ascii",
+                    "disable",
+                    "unit",
+                    "unit_scale",
+                    "dynamic_ncols",
+                    "smoothing",
+                    "bar_format",
+                    "initial",
+                    "position",
+                    "postfix",
+                    "unit_divisor",
+                    "write_bytes",
+                    "lock_args",
+                    "nrows",
+                    "colour",
+                    "color",
+                    "delay",
+                    "gui",
+                    "disable_default",
+                    "pos",
                }
                for key, value in kwargs.items():
                    if key in tqdm_kwargs:
                        filtered_kwargs[key] = value
-                
+
+                # Force-enable the progress bar — we're tracking progress ourselves,
+                # we don't need tqdm to render to a terminal, but we DO need
+                # self.n to be updated when update() is called.
+                filtered_kwargs["disable"] = False
+
                # Try to initialize with filtered kwargs, fall back to all kwargs if that fails
                try:
                    super().__init__(*args, **filtered_kwargs)
                except TypeError:
                    # If filtering failed, try with all kwargs (maybe tqdm version accepts them)
+                    kwargs["disable"] = False
                    super().__init__(*args, **kwargs)
-                
+
                self._tracker_filename = filename or "unknown"
-                
+
                with tracker._lock:
                    if filename:
                        tracker._current_filename = filename
                    tracker._active_tqdms[id(self)] = {
                        "filename": self._tracker_filename,
                    }
-            
+
            def update(self, n=1):
                result = super().update(n)

@@ -89,95 +122,97 @@ class HFProgressTracker:
                        filename = tracker._active_tqdms[id(self)]["filename"]
                        current = getattr(self, "n", 0)
                        total = getattr(self, "total", 0)
-                        
+
                        if total and total > 0:
                            # Always filter out non-byte progress bars (e.g., "Fetching 12 files")
                            # These cause crazy percentages because they're counting files, not bytes
                            if self._is_non_byte_progress(filename):
                                return result
-                            
+
                            # When model is cached, also filter out generation-related progress
                            if tracker.filter_non_downloads:
                                if not self._is_download_progress(filename):
                                    return result
-                            
+
                            # Update per-file tracking
                            tracker._file_sizes[filename] = total
                            tracker._file_downloaded[filename] = current
-                            
+
                            # Calculate totals across all files
                            tracker._total_size = sum(tracker._file_sizes.values())
                            tracker._total_downloaded = sum(tracker._file_downloaded.values())
-                            
+
                            # Only report progress once we have a meaningful total (at least 1MB)
                            # This avoids the "100% at 0MB" issue when small config
                            # files are counted before the real model files
                            MIN_TOTAL_BYTES = 1_000_000  # 1MB
                            if tracker._total_size < MIN_TOTAL_BYTES:
                                return result
-                            
+
                            # Call progress callback
                            if tracker.progress_callback:
-                                tracker.progress_callback(
-                                    tracker._total_downloaded,
-                                    tracker._total_size,
-                                    filename
-                                )
-                
+                                tracker.progress_callback(tracker._total_downloaded, tracker._total_size, filename)
+
                return result
-            
+
            def _is_non_byte_progress(self, filename: str) -> bool:
                """Check if this progress bar should be SKIPPED (returns True to skip).
-                
+
                We want to track byte-based progress bars. This method identifies
                progress bars that count files/items instead of bytes, which would
                cause crazy percentages if mixed with our byte counting.
-                
+
                Returns:
                    True = SKIP this bar (it's not byte-based)
                    False = TRACK this bar (it counts bytes)
                """
                if not filename:
                    return False
-                
+
                filename_lower = filename.lower()
-                
+
                # Skip "Fetching X files" - it counts files (total=12), not bytes
                # Don't skip "Downloading (incomplete total...)" - that IS byte-based
                skip_patterns = [
-                    'fetching',  # "Fetching 12 files" has total=12 files, not bytes
+                    "fetching",  # "Fetching 12 files" has total=12 files, not bytes
                ]
                return any(pattern in filename_lower for pattern in skip_patterns)
-            
+
            def _is_download_progress(self, filename: str) -> bool:
                """Check if this is a real file download progress bar vs internal processing."""
                if not filename or filename == "unknown":
                    return False
-                
+
                # Real downloads have file extensions
                download_extensions = [
-                    '.safetensors', '.bin', '.pt', '.pth',  # Model weights
-                    '.json', '.txt', '.py',  # Config files
-                    '.msgpack', '.h5',  # Other formats
+                    ".safetensors",
+                    ".bin",
+                    ".pt",
+                    ".pth",  # Model weights
+                    ".json",
+                    ".txt",
+                    ".py",  # Config files
+                    ".msgpack",
+                    ".h5",  # Other formats
                ]
-                
+
                filename_lower = filename.lower()
                has_extension = any(filename_lower.endswith(ext) for ext in download_extensions)
-                
+
                # Skip generation-related progress indicators
-                skip_patterns = ['segment', 'processing', 'generating', 'loading']
+                skip_patterns = ["segment", "processing", "generating", "loading"]
                has_skip_pattern = any(pattern in filename_lower for pattern in skip_patterns)
-                
+
                return has_extension and not has_skip_pattern
-            
+
            def close(self):
                with tracker._lock:
                    if id(self) in tracker._active_tqdms:
                        del tracker._active_tqdms[id(self)]
                return super().close()
-        
+
        return TrackedTqdm
-    
+
    @contextmanager
    def patch_download(self):
        """Context manager to patch tqdm for progress tracking."""
@@ -186,7 +221,7 @@ class HFProgressTracker:

            # Store original tqdm class
            self._original_tqdm_class = tqdm_module.tqdm
-            
+
            # Reset totals
            with self._lock:
                self._total_downloaded = 0
@@ -195,7 +230,7 @@ class HFProgressTracker:
                self._file_downloaded = {}
                self._current_filename = ""
                self._active_tqdms = {}
-            
+
            # Create our tracked tqdm class
            tracked_tqdm = self._create_tracked_tqdm_class()

@@ -207,13 +242,13 @@ class HFProgressTracker:
            if hasattr(tqdm_module, "auto") and hasattr(tqdm_module.auto, "tqdm"):
                self._original_tqdm_auto = tqdm_module.auto.tqdm
                tqdm_module.auto.tqdm = tracked_tqdm
-            
+
            # Patch in sys.modules to catch already-imported references
            # huggingface_hub uses: from tqdm.auto import tqdm as base_tqdm
            # So we need to patch both 'tqdm' and 'base_tqdm' attributes
            self._patched_modules = {}
-            tqdm_attr_names = ['tqdm', 'base_tqdm', 'old_tqdm']  # Various names used
-            
+            tqdm_attr_names = ["tqdm", "base_tqdm", "old_tqdm"]  # Various names used
+
            patched_count = 0
            for module_name in list(sys.modules.keys()):
                if "huggingface" in module_name or module_name.startswith("tqdm"):
@@ -224,10 +259,13 @@ class HFProgressTracker:
                                attr = getattr(module, attr_name)
                                # Only patch if it's a tqdm class (not already patched)
                                is_tqdm_class = (
-                                    attr is self._original_tqdm_class or 
-                                    (self._original_tqdm_auto and attr is self._original_tqdm_auto) or
-                                    (hasattr(attr, "__name__") and attr.__name__ == "tqdm" and 
-                                     hasattr(attr, "update"))  # tqdm classes have update method
+                                    attr is self._original_tqdm_class
+                                    or (self._original_tqdm_auto and attr is self._original_tqdm_auto)
+                                    or (
+                                        hasattr(attr, "__name__")
+                                        and attr.__name__ == "tqdm"
+                                        and hasattr(attr, "update")
+                                    )  # tqdm classes have update method
                                )
                                if is_tqdm_class:
                                    key = f"{module_name}.{attr_name}"
@@ -236,31 +274,33 @@ class HFProgressTracker:
                                    patched_count += 1
                    except (AttributeError, TypeError):
                        pass
-            
+
            # ALSO monkey-patch the update method on huggingface_hub's tqdm class
            # This is needed because the class was already defined at import time
            self._hf_tqdm_original_update = None
            try:
                from huggingface_hub.utils import tqdm as hf_tqdm_module
-                if hasattr(hf_tqdm_module, 'tqdm'):
+
+                if hasattr(hf_tqdm_module, "tqdm"):
                    hf_tqdm_class = hf_tqdm_module.tqdm
                    self._hf_tqdm_original_update = hf_tqdm_class.update
-                    
+
                    # Create a wrapper that calls our tracking
                    tracker = self  # Reference to HFProgressTracker instance
+
                    def patched_update(tqdm_self, n=1):
                        result = tracker._hf_tqdm_original_update(tqdm_self, n)
-                        
+
                        # Track this progress
                        with tracker._lock:
-                            desc = getattr(tqdm_self, 'desc', '') or ''
-                            current = getattr(tqdm_self, 'n', 0)
-                            total = getattr(tqdm_self, 'total', 0) or 0
-                            
+                            desc = getattr(tqdm_self, "desc", "") or ""
+                            current = getattr(tqdm_self, "n", 0)
+                            total = getattr(tqdm_self, "total", 0) or 0
+
                            # Skip non-byte progress bars
-                            if 'fetching' in desc.lower():
+                            if "fetching" in desc.lower():
                                return result
-                            
+
                            # Skip until we have a meaningful total (at least 1MB)
                            # This avoids the "100% at 0MB" issue when small config
                            # files are counted before the real model files
@@ -268,22 +308,22 @@ class HFProgressTracker:
                            if total >= MIN_TOTAL_BYTES:
                                tracker._total_downloaded = current
                                tracker._total_size = total
-                                
+
                                if tracker.progress_callback:
                                    tracker.progress_callback(current, total, desc)
-                        
+
                        return result
-                    
+
                    hf_tqdm_class.update = patched_update
                    patched_count += 1
-                    print(f"[HFProgressTracker] Monkey-patched huggingface_hub.utils.tqdm.tqdm.update")
+                    logger.debug("Monkey-patched huggingface_hub.utils.tqdm.tqdm.update")
            except (ImportError, AttributeError) as e:
-                print(f"[HFProgressTracker] Could not monkey-patch hf_tqdm: {e}")
-            
-            print(f"[HFProgressTracker] Patched {patched_count} tqdm references")
-            
+                logger.warning("Could not monkey-patch hf_tqdm: %s", e)
+
+            logger.debug("Patched %d tqdm references", patched_count)
+
            yield
-            
+
        except ImportError:
            # If tqdm not available, just yield without patching
            yield
@@ -292,11 +332,12 @@ class HFProgressTracker:
            if self._original_tqdm_class:
                try:
                    import tqdm as tqdm_module
+
                    tqdm_module.tqdm = self._original_tqdm_class
-                    
+
                    if self._original_tqdm_auto:
                        tqdm_module.auto.tqdm = self._original_tqdm_auto
-                    
+
                    # Restore patched modules
                    for key, (module, attr_name, original) in self._patched_modules.items():
                        try:
@@ -305,26 +346,28 @@ class HFProgressTracker:
                        except (AttributeError, TypeError):
                            pass
                    self._patched_modules = {}
-                    
+
                    # Restore hf_tqdm's original update method
                    if self._hf_tqdm_original_update:
                        try:
                            from huggingface_hub.utils import tqdm as hf_tqdm_module
-                            if hasattr(hf_tqdm_module, 'tqdm'):
+
+                            if hasattr(hf_tqdm_module, "tqdm"):
                                hf_tqdm_module.tqdm.update = self._hf_tqdm_original_update
                        except (ImportError, AttributeError):
                            pass
                        self._hf_tqdm_original_update = None
-                    
+
                except (ImportError, AttributeError):
                    pass


 def create_hf_progress_callback(model_name: str, progress_manager):
    """Create a progress callback for HuggingFace downloads."""
+
    def callback(downloaded: int, total: int, filename: str = ""):
        """Progress callback.
-        
+
        Note: We send updates even when total=0 (unknown) to provide feedback
        during the "incomplete total" phase of huggingface_hub downloads.
        The frontend handles total=0 gracefully.
@@ -336,4 +379,5 @@ def create_hf_progress_callback(model_name: str, progress_manager):
            filename=filename or "",
            status="downloading",
        )
+
    return callback
@@ -1,66 +0,0 @@
-"""
-Input validation utilities.
-"""
-
-from typing import Tuple, Optional
-from pathlib import Path
-
-
-def validate_text(text: str, max_length: int = 5000) -> Tuple[bool, Optional[str]]:
-    """
-    Validate text input.
-    
-    Args:
-        text: Text to validate
-        max_length: Maximum length
-        
-    Returns:
-        Tuple of (is_valid, error_message)
-    """
-    if not text or not text.strip():
-        return False, "Text cannot be empty"
-    
-    if len(text) > max_length:
-        return False, f"Text too long (maximum {max_length} characters)"
-    
-    return True, None
-
-
-def validate_language(language: str) -> Tuple[bool, Optional[str]]:
-    """
-    Validate language code.
-
-    Supported languages for Qwen3-TTS:
-    Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian
-
-    Args:
-        language: Language code
-
-    Returns:
-        Tuple of (is_valid, error_message)
-    """
-    valid_languages = ["zh", "en", "ja", "ko", "de", "fr", "ru", "pt", "es", "it"]
-    if language not in valid_languages:
-        return False, f"Invalid language (must be one of: {', '.join(valid_languages)})"
-
-    return True, None
-
-
-def validate_file_path(path: str) -> Tuple[bool, Optional[str]]:
-    """
-    Validate file path exists.
-    
-    Args:
-        path: File path
-        
-    Returns:
-        Tuple of (is_valid, error_message)
-    """
-    file_path = Path(path)
-    if not file_path.exists():
-        return False, f"File not found: {path}"
-    
-    if not file_path.is_file():
-        return False, f"Path is not a file: {path}"
-    
-    return True, None
@@ -6,13 +6,23 @@ from PyInstaller.utils.hooks import copy_metadata

 datas = []
 binaries = []
-hiddenimports = ['backend', 'backend.main', 'backend.config', 'backend.database', 'backend.models', 'backend.profiles', 'backend.history', 'backend.tts', 'backend.transcribe', 'backend.platform_detect', 'backend.backends', 'backend.backends.pytorch_backend', 'backend.utils.audio', 'backend.utils.cache', 'backend.utils.progress', 'backend.utils.hf_progress', 'backend.utils.validation', 'backend.cuda_download', 'backend.effects', 'backend.utils.effects', 'backend.versions', 'pedalboard', 'torch', 'transformers', 'fastapi', 'uvicorn', 'sqlalchemy', 'librosa', 'soundfile', 'qwen_tts', 'qwen_tts.inference', 'qwen_tts.inference.qwen3_tts_model', 'qwen_tts.inference.qwen3_tts_tokenizer', 'qwen_tts.core', 'qwen_tts.cli', 'pkg_resources.extern', 'backend.backends.mlx_backend', 'mlx', 'mlx.core', 'mlx.nn', 'mlx_audio', 'mlx_audio.tts', 'mlx_audio.stt']
+hiddenimports = ['backend', 'backend.main', 'backend.config', 'backend.database', 'backend.models', 'backend.profiles', 'backend.history', 'backend.tts', 'backend.transcribe', 'backend.platform_detect', 'backend.backends', 'backend.backends.pytorch_backend', 'backend.utils.audio', 'backend.utils.cache', 'backend.utils.progress', 'backend.utils.hf_progress', 'backend.utils.validation', 'backend.cuda_download', 'backend.effects', 'backend.utils.effects', 'backend.versions', 'pedalboard', 'chatterbox', 'chatterbox.tts_turbo', 'chatterbox.mtl_tts', 'backend.backends.chatterbox_backend', 'backend.backends.chatterbox_turbo_backend', 'backend.backends.luxtts_backend', 'zipvoice', 'zipvoice.luxvoice', 'torch', 'transformers', 'fastapi', 'uvicorn', 'sqlalchemy', 'librosa', 'soundfile', 'qwen_tts', 'qwen_tts.inference', 'qwen_tts.inference.qwen3_tts_model', 'qwen_tts.inference.qwen3_tts_tokenizer', 'qwen_tts.core', 'qwen_tts.cli', 'requests', 'pkg_resources.extern', 'backend.backends.mlx_backend', 'mlx', 'mlx.core', 'mlx.nn', 'mlx_audio', 'mlx_audio.tts', 'mlx_audio.stt']
 datas += collect_data_files('qwen_tts')
 datas += copy_metadata('qwen-tts')
+datas += copy_metadata('requests')
+datas += copy_metadata('transformers')
+datas += copy_metadata('huggingface-hub')
+datas += copy_metadata('tokenizers')
+datas += copy_metadata('safetensors')
+datas += copy_metadata('tqdm')
 hiddenimports += collect_submodules('qwen_tts')
 hiddenimports += collect_submodules('jaraco')
 hiddenimports += collect_submodules('mlx')
 hiddenimports += collect_submodules('mlx_audio')
+tmp_ret = collect_all('zipvoice')
+datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
+tmp_ret = collect_all('linacodec')
+datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
 tmp_ret = collect_all('mlx')
 datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
 tmp_ret = collect_all('mlx_audio')
@@ -1,3 +1,26 @@
-node_modules
-.mintlify
+# deps
+/node_modules
+
+# generated content
+.source
+
+# test & build
+/coverage
+/.next/
+/out/
+/build
+*.tsbuildinfo
+
+# misc
 .DS_Store
+*.pem
+/.pnp
+.pnp.js
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+
+# others
+.env*.local
+.vercel
+next-env.d.ts
@@ -1,192 +0,0 @@
-# Auto-Updater Documentation
-
-Voicebox includes automatic updates powered by Tauri's updater plugin. This document explains how it works for both users and developers.
-
-## 1. Generate Signing Keys
-
-Run this command to generate your signing keypair:
-
-```bash
-cd tauri && bun tauri signer generate -w ~/.tauri/voicebox.key
-```
-
-This creates:
- **Private key**: `~/.tauri/voicebox.key` (keep this secret!)
- **Public key**: `~/.tauri/voicebox.key.pub`
-
-## 2. Update Configuration
-
-Copy the content from `~/.tauri/voicebox.key.pub` and replace the placeholder in `tauri/src-tauri/tauri.conf.json`:
-
-```json
-{
-  "plugins": {
-    "updater": {
-      "pubkey": "PASTE_PUBLIC_KEY_CONTENT_HERE",
-      "endpoints": [
-        "https://github.com/YOUR_USERNAME/voicebox/releases/latest/download/latest.json"
-      ]
-    }
-  }
-}
-```
-
-Update the endpoint URL with your actual GitHub username/organization.
-
-## 3. Building with Signatures
-
-When building releases, set these environment variables:
-
-**macOS/Linux:**
-```bash
-export TAURI_SIGNING_PRIVATE_KEY="$(cat ~/.tauri/voicebox.key)"
-export TAURI_SIGNING_PRIVATE_KEY_PASSWORD=""
-bun run build
-```
-
-**Windows PowerShell:**
-```powershell
-$env:TAURI_SIGNING_PRIVATE_KEY = Get-Content ~/.tauri/voicebox.key -Raw
-$env:TAURI_SIGNING_PRIVATE_KEY_PASSWORD = ""
-bun run build
-```
-
-## 4. GitHub Release Setup
-
-When you create a GitHub release, the build process will generate:
- Installers for each platform
- `.sig` signature files
- `latest.json` update manifest
-
-### Manual Release Process
-
-1. Build the app with signing keys set
-2. Create a new GitHub release
-3. Upload all files from `tauri/src-tauri/target/release/bundle/`
-4. Create `latest.json` in your release assets:
-
-```json
-{
-  "version": "0.2.0",
-  "notes": "Bug fixes and improvements",
-  "pub_date": "2026-01-25T12:00:00Z",
-  "platforms": {
-    "darwin-aarch64": {
-      "signature": "CONTENT_FROM_.app.tar.gz.sig",
-      "url": "https://github.com/YOUR_USERNAME/voicebox/releases/download/v0.2.0/voicebox_0.2.0_aarch64.dmg"
-    },
-    "darwin-x86_64": {
-      "signature": "CONTENT_FROM_.app.tar.gz.sig",
-      "url": "https://github.com/YOUR_USERNAME/voicebox/releases/download/v0.2.0/voicebox_0.2.0_x64.dmg"
-    },
-    "linux-x86_64": {
-      "signature": "CONTENT_FROM_.AppImage.sig",
-      "url": "https://github.com/YOUR_USERNAME/voicebox/releases/download/v0.2.0/voicebox_0.2.0_amd64.AppImage"
-    },
-    "windows-x86_64": {
-      "signature": "CONTENT_FROM_.msi.sig",
-      "url": "https://github.com/YOUR_USERNAME/voicebox/releases/download/v0.2.0/voicebox_0.2.0_x64_en-US.msi"
-    }
-  }
-}
-```
-
-### Automated GitHub Actions (Recommended)
-
-Create `.github/workflows/release.yml`:
-
-```yaml
-name: Release
-
-on:
-  push:
-    tags:
-      - 'v*'
-
-jobs:
-  release:
-    strategy:
-      matrix:
-        platform: [macos-latest, ubuntu-22.04, windows-latest]
-
-    runs-on: ${{ matrix.platform }}
-
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Setup Bun
-        uses: oven-sh/setup-bun@v1
-
-      - name: Install Rust
-        uses: dtolnay/rust-toolchain@stable
-
-      - name: Install dependencies (Ubuntu)
-        if: matrix.platform == 'ubuntu-22.04'
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y libwebkit2gtk-4.1-dev libappindicator3-dev librsvg2-dev patchelf
-
-      - name: Install dependencies
-        run: bun install
-
-      - name: Build
-        env:
-          TAURI_SIGNING_PRIVATE_KEY: ${{ secrets.TAURI_SIGNING_PRIVATE_KEY }}
-          TAURI_SIGNING_PRIVATE_KEY_PASSWORD: ${{ secrets.TAURI_SIGNING_PRIVATE_KEY_PASSWORD }}
-        run: bun run build
-
-      - name: Upload Release
-        uses: softprops/action-gh-release@v1
-        with:
-          files: tauri/src-tauri/target/release/bundle/**/*
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-```
-
-Add your private key to GitHub secrets:
- Go to Settings → Secrets and variables → Actions
- Add `TAURI_SIGNING_PRIVATE_KEY` with the content of `~/.tauri/voicebox.key`
- Add `TAURI_SIGNING_PRIVATE_KEY_PASSWORD` (empty string if no password)
-
-## Frontend Integration
-
-The frontend integration is complete with automatic update notifications and manual update checks:
-
- **Update Notification Banner** - Appears automatically when updates are available
- **Settings Panel** - Manual "Check for Updates" button in Settings tab
- **Update Hook** - React hook handles all update operations
-
-See `docs/AUTOUPDATER_QUICKSTART.md` for a quick setup guide.
-
-## Security Notes
-
- Never commit your private key to version control
- Store private keys securely (use GitHub secrets for CI/CD)
- The public key in `tauri.conf.json` is safe to commit
- Updates are cryptographically verified before installation
- HTTP endpoints are blocked by default (HTTPS only)
-
-## Testing Updates
-
-1. Build version 0.1.0 and install it
-2. Update version in `tauri.conf.json` to 0.2.0
-3. Build version 0.2.0 with signatures
-4. Create a local server or GitHub release with `latest.json`
-5. Run version 0.1.0 and trigger update check
-6. Verify update downloads and installs correctly
-
-## Troubleshooting
-
-**"Invalid signature" error:**
- Verify public key matches the private key used to sign
- Ensure signature files (.sig) are uploaded correctly
-
-**"No update available" when one exists:**
- Check endpoint URL is correct
- Verify `latest.json` format matches specification
- Ensure version in latest.json is higher than current version
-
-**Build fails with signing:**
- Confirm environment variables are set correctly
- Check private key file exists and is readable
- Verify private key format (should start with `dW50cnVzdGVkIGNvbW1lbnQ6`)
@@ -1,116 +0,0 @@
-# Autoupdater Quick Start
-
-The Tauri v2 autoupdater has been fully configured and integrated. Follow these steps to activate it.
-
-## What's Already Done
-
-✅ Rust plugin installed and initialized
-✅ Tauri configuration set up with updater settings
-✅ Permissions granted for update operations
-✅ GitHub Actions workflow updated with signing support
-✅ Frontend components created and integrated
-✅ Update notifications on app startup
-✅ Manual update check in Settings tab
-
-## Required Steps (5 minutes)
-
-### 1. Generate Signing Keys
-
-```bash
-bun run generate:keys
-```
-
-This creates:
- Private key: `~/.tauri/voicebox.key` (keep secret!)
- Public key: `~/.tauri/voicebox.key.pub` (safe to share)
-
-### 2. Update Tauri Config
-
-Open `tauri/src-tauri/tauri.conf.json` and:
-
-1. Replace `"REPLACE_WITH_YOUR_PUBLIC_KEY"` with the content from `~/.tauri/voicebox.key.pub`
-2. Update the endpoint URL with your GitHub username:
-   ```json
-   "endpoints": [
-     "https://github.com/YOUR_USERNAME/voicebox/releases/latest/download/latest.json"
-   ]
-   ```
-
-### 3. Add GitHub Secrets
-
-Go to your repo Settings → Secrets and variables → Actions:
-
-1. Add `TAURI_SIGNING_PRIVATE_KEY`:
-   ```bash
-   cat ~/.tauri/voicebox.key
-   ```
-   Copy the entire output and paste as the secret value
-
-2. Add `TAURI_SIGNING_PRIVATE_KEY_PASSWORD`:
-   Leave empty (or add your password if you set one)
-
-### 4. Test the Setup
-
-To test locally before creating a release:
-
-```bash
-bun run build:release
-```
-
-This will verify your keys are set up correctly.
-
-## How It Works
-
-### For Users
-1. App checks for updates on startup (only in Tauri builds)
-2. If an update is available, a banner appears at the top
-3. Users can click "Install Now" to download and install
-4. App restarts automatically after installation
-
-### For Developers
-1. Create a new git tag: `git tag v0.2.0 && git push --tags`
-2. GitHub Actions builds signed releases for all platforms
-3. Uploads installers and generates `latest.json` manifest
-4. Users running older versions will be notified automatically
-
-## UI Components
-
-### Update Notification Banner
- Shows at top of app when update is available
- Appears automatically on startup
- Displays download/install progress
-
-### Settings Panel
- Located in Settings tab
- Shows current version
- Manual "Check for Updates" button
- Update status and progress
-
-## Troubleshooting
-
-**"Public key not configured"**
- Make sure you copied the entire content from `voicebox.key.pub`
- The key should start with `dW50cnVzdGVkIGNvbW1lbnQ6`
-
-**"Failed to check for updates"**
- Endpoint URL might be incorrect
- No releases published yet (expected for first setup)
-
-**Build fails with signing error**
- Check that GitHub secrets are set correctly
- Verify private key file exists at `~/.tauri/voicebox.key`
-
-## Next Release Workflow
-
-1. Update version in `tauri/src-tauri/tauri.conf.json`
-2. Commit changes
-3. Create and push tag: `git tag v0.2.0 && git push --tags`
-4. GitHub Actions will automatically build and create a draft release
-5. Review the release and publish it
-6. Users will be notified of the update
-
-## See Also
-
- Full documentation: `docs/AUTOUPDATER.md`
- Build script: `scripts/prepare-release.sh`
- GitHub workflow: `.github/workflows/release.yml`
@@ -0,0 +1,87 @@
+# Documentation Migration: Mintlify → Fumadocs
+
+This document summarizes the migration of documentation from `/docs` (Mintlify) to `/docs2` (Fumadocs).
+
+## What Was Done
+
+### 1. Files Copied
+- ✅ All 29 MDX files from `/docs` folders (overview, api, developer, plans)
+- ✅ All 4 root-level markdown files (AUTOUPDATER.md, AUTOUPDATER_QUICKSTART.md, TROUBLESHOOTING.md, README.md)
+- ✅ All images (3 webp files) → `public/images/`
+- ✅ All logo files (2 png files) → `public/logo/`
+
+### 2. Component Migration
+Created compatibility layer in `components/mintlify-compat.tsx` that maps Mintlify components to Fumadocs equivalents:
+
+- `<Frame>` → Simple div wrapper (images are zoomable by default in Fumadocs)
+- `<CardGroup>` → `<Cards>` (Fumadocs component)
+- `<Card>` → `<Card>` (with icon string → Lucide icon mapping)
+- `<Steps>` / `<Step>` → Direct mapping to Fumadocs components
+- `<Tip>`, `<Note>`, `<Info>` → `<Callout type="info">`
+- `<Warning>` → `<Callout type="warn">`
+- `<Danger>` → `<Callout type="error">`
+- `<AccordionGroup>` / `<Accordion>` → HTML `<details>` / `<summary>` elements
+
+### 3. Navigation Structure
+Created `meta.json` files for each folder:
+- `content/docs/meta.json` - Root documentation
+- `content/docs/overview/meta.json` - Overview pages
+- `content/docs/api/meta.json` - API reference
+- `content/docs/developer/meta.json` - Developer docs
+- `content/docs/plans/meta.json` - Plans/roadmap
+
+### 4. Link Fixes
+- Fixed incorrect `/guides/...` paths → `/overview/...`
+- All internal links now use correct paths
+
+### 5. Branding
+- Updated `lib/layout.shared.tsx` to use "Voicebox" as the nav title
+
+## File Structure
+
+```
+docs2/
+├── components/
+│   └── mintlify-compat.tsx    # Mintlify → Fumadocs component mappings
+├── content/docs/
+│   ├── meta.json              # Root navigation
+│   ├── overview/              # 12 MDX files
+│   ├── api/                   # 5 MDX files
+│   ├── developer/             # 12 MDX files
+│   ├── plans/                 # 4 MD files
+│   └── *.md                   # 4 root markdown files
+├── public/
+│   ├── images/                # 3 webp files
+│   └── logo/                  # 2 png files
+└── mdx-components.tsx         # MDX component configuration
+```
+
+## Icon Mapping
+
+The following icon strings are mapped to Lucide icons:
+- `microphone` → Mic
+- `film` → Film
+- `code` → Code
+- `shield` → Shield
+- `download` → Download
+- `rocket` → Rocket
+- `apple` → Apple
+- `windows` → Windows
+- `server` → Server
+- `user` → User
+- `waveform` → Waveform
+
+## Next Steps
+
+1. **Test the build**: Run `npm run build` (requires Node.js >= 20.9.0)
+2. **Start dev server**: Run `npm run dev` to preview
+3. **Customize styling**: Update `app/global.css` if needed
+4. **Add more icons**: Extend `iconMap` in `mintlify-compat.tsx` as needed
+5. **Review navigation**: Adjust `meta.json` files to customize page order
+
+## Notes
+
+- Image paths (`/images/...`) work as-is since Next.js serves from `public/`
+- All Mintlify components are now compatible with Fumadocs
+- Navigation structure follows Fumadocs conventions
+- No breaking changes to content - all MDX files work with compatibility layer
@@ -1,70 +0,0 @@
-# Accessibility: screen reader and keyboard improvements
-
-## Summary
-
-Improvements to support screen reader and keyboard users across the main app surfaces: audio player, generation UI, voice selection, history, voices tab, model management, server tab, and stories.
-
-**Tested with NVDA and Narrator on Windows.**
-
---
-
-## What changed
-
-### Audio player (after generating audio)
-
- **Play/Pause, Loop, Mute, Close** – `aria-label` added so each control is announced (e.g. "Play", "Pause", "Loop", "Mute", "Close player").
- **Playback position slider** – `aria-label="Playback position"` and `aria-valuetext` with current/total time (e.g. "0:30 of 2:15").
- **Volume** – Wrapped in a labelled group; volume slider has an associated screen-reader-only label and `aria-valuetext` for the level (e.g. "Volume level, 75%").
-
-### Generation UI (text box and voice choice)
-
- **Generate speech** (submit) and **Fine-tune instructions** (sliders) – Icon buttons now have `aria-label` (and state for fine-tune, e.g. "Fine-tune instructions, on").
-
-### Voice selection (cards on Generate screen)
-
- Each **voice card** is focusable (`tabIndex={0}`), has `role="button"`, and an `aria-label` (e.g. "Prashant, en. Select as voice for generation.") with `aria-pressed` when selected.
- **Enter/Space** on the card selects that voice; tab order is card → Export/Edit/Delete.
-
-### History list (generated samples)
-
- Each **sample row** is focusable with `role="button"` and an `aria-label` (e.g. "Sample from [profile], [duration], [date]. Press Enter to play."); **Enter/Space** plays or restarts.
- **Transcript textarea** has `aria-label` (e.g. "Transcript for sample from [profile], [duration]") so when you focus on the text area, the sample is announced in context.
-
-### Voices tab (table)
-
- Each **voice row** is focusable with `role="button"` and an `aria-label` (e.g. "[Name], [language], [N] generations, [N] samples. Press Enter to edit."); **Enter/Space** opens edit (except when focus is in a control).
- **Actions** dropdown trigger has `aria-label="Actions for [profile name]"`.
-
-### Model management
-
- Each **model row** is a focusable region (`tabIndex={0}`, `role="group"`) with an `aria-label` (e.g. "[Model name], [status], [size]. Use Tab to reach Download or Delete.").
- **Download** and **Delete** (and Downloading) buttons have `aria-label` (e.g. "Download [name]", "Delete [name]").
-
-### Server tab (panels)
-
- **Server Connection**, **Server Status**, and **App Updates** cards are landmarks: `role="region"`, `aria-label`, and `tabIndex={0}` so each panel is focusable and announced (e.g. "Server Connection", "Server Status", "App Updates").
-
-### Stories list
-
- Each **story row** is a focusable control (`role="button"`, `tabIndex={0}`) with `aria-label` (e.g. "Story [name], [N] items, [date]. Press Enter to select."); **Enter/Space** selects the story. Actions button has `aria-label="Actions for [story name]"`.
-
-### Other controls
-
- **Story list** – Actions (⋮) button: `aria-label="Actions for [story name]"`.
- **Story track editor** – Play/Pause, Stop, Split, Duplicate, Delete, Zoom in/out: `aria-label` on all icon buttons.
- **Voice profile samples** (SampleList, AudioSampleUpload, AudioSampleRecording, AudioSampleSystem) – Play/Pause and Stop: `aria-label` (e.g. "Play sample", "Pause", "Stop playback").
- **SampleList** mini sample player – Seek slider has `aria-label="Sample playback position"` and `aria-valuetext` for time.
-
---
-
-## Testing
-
- **Screen readers:** Tested with **NVDA** and **Narrator** on Windows.
- **Keyboard:** Tab order and Enter/Space activation verified for focusable rows and buttons.
-
---
-
-## Tech note
-
- React + TypeScript; Radix UI primitives; labels added via `aria-label`, `aria-labelledby`, `aria-valuetext`, and `role`/`tabIndex` where needed.
- No new dependencies.
@@ -1,64 +1,45 @@
-# Voicebox Documentation
+# fumadocs-ui-template

-This directory contains the documentation for Voicebox, built with [Mintlify](https://mintlify.com).
+This is a Next.js application generated with
+[Create Fumadocs](https://github.com/fuma-nama/fumadocs).

-## Development
-
-### Prerequisites
-
-Install Mintlify globally using bun:
+Run development server:

 ```bash
-bun add -g mintlify
+npm run dev
+# or
+pnpm dev
+# or
+yarn dev
 ```

-Or use the helper script:
+Open http://localhost:3000 with your browser to see the result.

-```bash
-bun run install:mintlify
-```
+## Explore

-### Running Locally
+In the project, you can see:

-```bash
-bun run dev
-```
+- `lib/source.ts`: Code for content source adapter, [`loader()`](https://fumadocs.dev/docs/headless/source-api) provides the interface to access your content.
+- `lib/layout.shared.tsx`: Shared options for layouts, optional but preferred to keep.

-This will start the Mintlify dev server.
+| Route                     | Description                                            |
+| ------------------------- | ------------------------------------------------------ |
+| `app/(home)`              | The route group for your landing page and other pages. |
+| `app/docs`                | The documentation layout and pages.                    |
+| `app/api/search/route.ts` | The Route Handler for search.                          |

-The docs will be available at `http://localhost:3000`
+### Fumadocs MDX

-### Structure
+A `source.config.ts` config file has been included, you can customise different options like frontmatter schema.

-```
-docs/
-├── mint.json           # Mintlify configuration
-├── custom.css          # Custom styles
-├── overview/           # Getting started & feature docs
-├── guides/             # User guides
-├── api/                # API reference
-├── development/        # Developer documentation
-├── logo/               # Logo assets
-└── public/             # Static assets
-```
+Read the [Introduction](https://fumadocs.dev/docs/mdx) for further details.

-### Writing Docs
+## Learn More

- Use `.mdx` files for all documentation pages
- Follow the existing structure in `mint.json` for navigation
- Use Mintlify components for enhanced formatting (Card, CardGroup, Accordion, etc.)
- Reference the [Mintlify documentation](https://mintlify.com/docs) for available components
+To learn more about Next.js and Fumadocs, take a look at the following
+resources:

-## Deployment
-
-Docs are automatically deployed when changes are pushed to the main branch.
-
-To manually deploy:
-
-```bash
-mintlify deploy
-```
-
-## Contributing
-
-See [CONTRIBUTING.md](../CONTRIBUTING.md) for contribution guidelines.
+- [Next.js Documentation](https://nextjs.org/docs) - learn about Next.js
+  features and API.
+- [Learn Next.js](https://nextjs.org/learn) - an interactive Next.js tutorial.
+- [Fumadocs](https://fumadocs.dev) - learn about Fumadocs
@@ -1,55 +0,0 @@
---
-title: "Authentication"
-description: "API authentication and security"
---
-
-## Current Status
-
-<Warning>
-  Authentication is not currently implemented in Voicebox. The API is intended for local use only.
-</Warning>
-
-## Local Usage
-
-For local development and usage:
- API runs on `localhost:17493`
- No authentication required
- Access restricted to local machine
-
-## Future Implementation
-
-Authentication will be added in a future release for:
- Remote deployments
- Multi-user access
- Production environments
-
-Planned authentication methods:
- API keys
- OAuth 2.0
- JWT tokens
-
-## Security Best Practices
-
-Until authentication is implemented:
-
-<CardGroup cols={2}>
-  <Card title="Use VPN" icon="shield">
-    Use WireGuard or Tailscale for remote access
-  </Card>
-  <Card title="Reverse Proxy" icon="server">
-    Run behind nginx with basic auth
-  </Card>
-  <Card title="Firewall" icon="fire">
-    Restrict access to trusted IPs only
-  </Card>
-  <Card title="Local Only" icon="laptop">
-    Don't expose to public internet
-  </Card>
-</CardGroup>
-
-## Coming Soon
-
- API key management
- User accounts
- Rate limiting
- Access control
@@ -1,119 +0,0 @@
---
-title: "Generation API"
-description: "Generate speech from text"
---
-
-## Generate Speech
-
-```http
-POST /generate
-```
-
-**Request:**
-```json
-{
-  "text": "Hello world",
-  "profile_id": "abc123",
-  "language": "en"
-}
-```
-
-**Response:**
-```json
-{
-  "id": "gen123",
-  "text": "Hello world",
-  "profile_id": "abc123",
-  "language": "en",
-  "audio_url": "/audio/gen123.wav",
-  "duration": 2.3,
-  "created_at": "2024-01-29T12:00:00Z"
-}
-```
-
-## List History
-
-```http
-GET /history
-```
-
-**Query Parameters:**
- `profile_id` (optional) - Filter by voice profile
- `limit` (optional) - Number of results (default: 50)
- `offset` (optional) - Pagination offset
-
-**Response:**
-```json
-{
-  "generations": [
-    {
-      "id": "gen123",
-      "text": "Hello world",
-      "profile_id": "abc123",
-      "duration": 2.3,
-      "created_at": "2024-01-29T12:00:00Z"
-    }
-  ],
-  "total": 100
-}
-```
-
-## Get Generation
-
-```http
-GET /history/{id}
-```
-
-**Response:**
-```json
-{
-  "id": "gen123",
-  "text": "Hello world",
-  "profile_id": "abc123",
-  "language": "en",
-  "audio_url": "/audio/gen123.wav",
-  "duration": 2.3,
-  "created_at": "2024-01-29T12:00:00Z"
-}
-```
-
-## Delete Generation
-
-```http
-DELETE /history/{id}
-```
-
-**Response:**
-```json
-{
-  "success": true
-}
-```
-
-## TypeScript Example
-
-```typescript
-import { VoiceboxClient } from '@/lib/api'
-
-const client = new VoiceboxClient({
-  baseUrl: 'http://localhost:17493'
-})
-
-// Generate speech
-const generation = await client.generate({
-  text: 'Hello world',
-  profile_id: 'abc123',
-  language: 'en'
-})
-
-// Get audio URL
-const audioUrl = generation.audio_url
-
-// List history
-const history = await client.listHistory({
-  profile_id: 'abc123',
-  limit: 20
-})
-```
-
-For full API documentation, visit `http://localhost:17493/docs` when the server is running.
@@ -1,219 +0,0 @@
---
-title: "API Overview"
-description: "Integrate voice synthesis into your applications with the Voicebox REST API"
---
-
-## Introduction
-
-Voicebox exposes a full REST API that allows you to integrate voice synthesis into your own applications. The API runs on `http://localhost:17493` by default.
-
-<Card title="Interactive API Docs" icon="book" href="http://localhost:17493/docs">
-  When Voicebox is running, visit the auto-generated API documentation at `http://localhost:17493/docs`
-</Card>
-
-## Base URL
-
-```
-http://localhost:17493
-```
-
-For remote deployments, replace `localhost` with your server's IP or hostname.
-
-## Authentication
-
-<Note>
-  Currently, the API does not require authentication for local development. Authentication will be added in a future release for production deployments.
-</Note>
-
-## Quick Example
-
-Here's a simple example of generating speech:
-
-```bash
-# Generate speech
-curl -X POST http://localhost:17493/generate \
-  -H "Content-Type: application/json" \
-  -d '{
-    "text": "Hello world",
-    "profile_id": "abc123",
-    "language": "en"
-  }'
-```
-
-## API Endpoints
-
-The Voicebox API is organized into several categories:
-
-<CardGroup cols={2}>
-  <Card title="Voice Profiles" icon="user" href="/api/voice-profiles">
-    Create, list, update, and delete voice profiles
-  </Card>
-  <Card title="Generation" icon="waveform" href="/api/generation">
-    Generate speech from text using voice profiles
-  </Card>
-  <Card title="Recordings" icon="microphone" href="/api/recordings">
-    Record and transcribe audio
-  </Card>
-  <Card title="Stories" icon="film">
-    Create and manage multi-voice stories (coming soon)
-  </Card>
-</CardGroup>
-
-## Core Endpoints
-
-### Voice Profiles
-
-```http
-GET    /profiles              # List all profiles
-POST   /profiles              # Create a new profile
-GET    /profiles/{id}         # Get profile details
-PUT    /profiles/{id}         # Update a profile
-DELETE /profiles/{id}         # Delete a profile
-POST   /profiles/{id}/samples # Add voice sample
-```
-
-### Generation
-
-```http
-POST   /generate              # Generate speech
-GET    /history               # List generation history
-GET    /history/{id}          # Get generation details
-DELETE /history/{id}          # Delete from history
-```
-
-### Recordings
-
-```http
-POST   /recordings            # Start recording
-POST   /recordings/stop       # Stop recording
-POST   /transcribe            # Transcribe audio
-```
-
-## Response Format
-
-All API responses follow a consistent JSON format:
-
-```json
-{
-  "success": true,
-  "data": {
-    // Response data
-  },
-  "error": null
-}
-```
-
-Error responses:
-
-```json
-{
-  "success": false,
-  "data": null,
-  "error": {
-    "message": "Error description",
-    "code": "ERROR_CODE"
-  }
-}
-```
-
-## Data Models
-
-### Voice Profile
-
-```json
-{
-  "id": "abc123",
-  "name": "John Smith",
-  "language": "en",
-  "description": "Professional narrator voice",
-  "created_at": "2024-01-29T12:00:00Z",
-  "samples": [
-    {
-      "id": "sample123",
-      "audio_path": "/path/to/sample.wav",
-      "duration": 15.5
-    }
-  ]
-}
-```
-
-### Generation
-
-```json
-{
-  "id": "gen123",
-  "text": "Hello world",
-  "profile_id": "abc123",
-  "language": "en",
-  "audio_path": "/path/to/output.wav",
-  "duration": 2.3,
-  "created_at": "2024-01-29T12:00:00Z"
-}
-```
-
-## TypeScript Client
-
-Voicebox provides an auto-generated TypeScript client with full type safety:
-
-```typescript
-import { VoiceboxClient } from '@/lib/api'
-
-const client = new VoiceboxClient({
-  baseUrl: 'http://localhost:17493'
-})
-
-// Create a profile
-const profile = await client.createProfile({
-  name: 'John Smith',
-  language: 'en'
-})
-
-// Generate speech
-const generation = await client.generate({
-  text: 'Hello world',
-  profile_id: profile.id,
-  language: 'en'
-})
-```
-
-The client is automatically generated from the OpenAPI schema. See [Development Setup](/development/setup#generate-openapi-client) for details.
-
-## Rate Limiting
-
-<Info>
-  Currently, there are no rate limits for local usage. Rate limiting will be added in a future release for production deployments.
-</Info>
-
-## WebSocket Support
-
-<Note>
-  Real-time streaming generation via WebSockets is planned for a future release.
-</Note>
-
-## Use Cases
-
-<CardGroup cols={2}>
-  <Card title="Game Development" icon="gamepad">
-    Generate dynamic dialogue for NPCs and characters
-  </Card>
-  <Card title="Content Creation" icon="video">
-    Automate voiceovers for videos and podcasts
-  </Card>
-  <Card title="Accessibility" icon="universal-access">
-    Build text-to-speech tools for visually impaired users
-  </Card>
-  <Card title="Voice Assistants" icon="robot">
-    Create custom voice interfaces
-  </Card>
-</CardGroup>
-
-## Next Steps
-
-<CardGroup cols={2}>
-  <Card title="Voice Profiles API" icon="user" href="/api/voice-profiles">
-    Learn how to manage voice profiles
-  </Card>
-  <Card title="Generation API" icon="waveform" href="/api/generation">
-    Generate speech from text
-  </Card>
-</CardGroup>
@@ -1,95 +0,0 @@
---
-title: "Recordings API"
-description: "Record and transcribe audio"
---
-
-## Start Recording
-
-```http
-POST /recordings/start
-```
-
-**Request:**
-```json
-{
-  "source": "microphone"
-}
-```
-
-**Response:**
-```json
-{
-  "recording_id": "rec123",
-  "status": "recording"
-}
-```
-
-## Stop Recording
-
-```http
-POST /recordings/stop
-```
-
-**Request:**
-```json
-{
-  "recording_id": "rec123"
-}
-```
-
-**Response:**
-```json
-{
-  "recording_id": "rec123",
-  "audio_url": "/audio/rec123.wav",
-  "duration": 15.5
-}
-```
-
-## Transcribe Audio
-
-```http
-POST /transcribe
-```
-
-**Request:** (multipart/form-data)
-```
-audio: <file>
-language: "en" (optional)
-```
-
-**Response:**
-```json
-{
-  "text": "Transcribed speech text here",
-  "language": "en",
-  "duration": 15.5,
-  "confidence": 0.95
-}
-```
-
-## TypeScript Example
-
-```typescript
-import { VoiceboxClient } from '@/lib/api'
-
-const client = new VoiceboxClient({
-  baseUrl: 'http://localhost:17493'
-})
-
-// Start recording
-const recording = await client.startRecording({
-  source: 'microphone'
-})
-
-// ... record audio ...
-
-// Stop recording
-const result = await client.stopRecording(recording.id)
-
-// Transcribe
-const transcription = await client.transcribe(audioFile, 'en')
-console.log(transcription.text)
-```
-
-For full API documentation, visit `http://localhost:17493/docs` when the server is running.
@@ -1,149 +0,0 @@
---
-title: "Voice Profiles API"
-description: "Manage voice profiles programmatically"
---
-
-## Endpoints
-
-### List Profiles
-
-```http
-GET /profiles
-```
-
-**Response:**
-```json
-{
-  "profiles": [
-    {
-      "id": "abc123",
-      "name": "John Smith",
-      "language": "en",
-      "description": "Professional narrator",
-      "created_at": "2024-01-29T12:00:00Z",
-      "sample_count": 2
-    }
-  ]
-}
-```
-
-### Get Profile
-
-```http
-GET /profiles/{id}
-```
-
-**Response:**
-```json
-{
-  "id": "abc123",
-  "name": "John Smith",
-  "language": "en",
-  "description": "Professional narrator",
-  "created_at": "2024-01-29T12:00:00Z",
-  "samples": [
-    {
-      "id": "sample123",
-      "duration": 15.5,
-      "created_at": "2024-01-29T12:00:00Z"
-    }
-  ]
-}
-```
-
-### Create Profile
-
-```http
-POST /profiles
-```
-
-**Request:**
-```json
-{
-  "name": "John Smith",
-  "language": "en",
-  "description": "Professional narrator"
-}
-```
-
-**Response:**
-```json
-{
-  "id": "abc123",
-  "name": "John Smith",
-  "language": "en",
-  "description": "Professional narrator",
-  "created_at": "2024-01-29T12:00:00Z"
-}
-```
-
-### Update Profile
-
-```http
-PUT /profiles/{id}
-```
-
-**Request:**
-```json
-{
-  "name": "Updated Name",
-  "description": "Updated description"
-}
-```
-
-### Delete Profile
-
-```http
-DELETE /profiles/{id}
-```
-
-**Response:**
-```json
-{
-  "success": true
-}
-```
-
-### Add Voice Sample
-
-```http
-POST /profiles/{id}/samples
-```
-
-**Request:** (multipart/form-data)
-```
-audio: <file>
-```
-
-**Response:**
-```json
-{
-  "sample_id": "sample123",
-  "duration": 15.5
-}
-```
-
-## TypeScript Example
-
-```typescript
-import { VoiceboxClient } from '@/lib/api'
-
-const client = new VoiceboxClient({
-  baseUrl: 'http://localhost:17493'
-})
-
-// Create profile
-const profile = await client.createProfile({
-  name: 'John Smith',
-  language: 'en',
-  description: 'Professional narrator'
-})
-
-// Add sample
-await client.addSample(profile.id, audioFile)
-
-// List all profiles
-const profiles = await client.listProfiles()
-```
-
-For full API documentation, visit `http://localhost:17493/docs` when the server is running.
@@ -0,0 +1,6 @@
+import { HomeLayout } from 'fumadocs-ui/layouts/home';
+import { baseOptions } from '@/lib/layout.shared';
+
+export default function Layout({ children }: LayoutProps<'/'>) {
+  return <HomeLayout {...baseOptions()}>{children}</HomeLayout>;
+}
@@ -0,0 +1,5 @@
+import { redirect } from 'next/navigation';
+
+export default function HomePage() {
+  redirect('/docs');
+}
@@ -0,0 +1,7 @@
+import { source } from '@/lib/source';
+import { createFromSource } from 'fumadocs-core/search/server';
+
+export const { GET } = createFromSource(source, {
+  // https://docs.orama.com/docs/orama-js/supported-languages
+  language: 'english',
+});
@@ -0,0 +1,74 @@
+import { createRelativeLink } from 'fumadocs-ui/mdx';
+import { DocsBody, DocsDescription, DocsPage, DocsTitle } from 'fumadocs-ui/page';
+import type { Metadata } from 'next';
+import { notFound } from 'next/navigation';
+import { MarkdownCopyButton, ViewOptionsPopover } from '@/components/ai/page-actions';
+import { APIPage } from '@/components/api-page';
+import { getPageImage, source } from '@/lib/source';
+import { getMDXComponents } from '@/mdx-components';
+
+export default async function Page(props: PageProps<'/docs/[[...slug]]'>) {
+  const params = await props.params;
+  const page = source.getPage(params.slug);
+  if (!page) notFound();
+
+  const MDX = page.data.body;
+  const markdownUrl = `${page.url}.mdx`;
+  const githubUrl = `https://github.com/jamiepine/voicebox/blob/main/docs/content/docs/${page.path}`;
+
+  return (
+    <DocsPage
+      toc={page.data.toc}
+      full={page.data.full}
+      editOnGithub={{
+        owner: 'jamiepine',
+        repo: 'voicebox',
+        sha: 'main',
+        path: `docs/content/docs/${page.path}`,
+      }}
+      lastUpdate={page.data.lastModified}
+    >
+      <DocsTitle>{page.data.title}</DocsTitle>
+      <DocsDescription className="mb-0">{page.data.description}</DocsDescription>
+      <div className="flex flex-row gap-2 items-center">
+        <MarkdownCopyButton markdownUrl={markdownUrl} />
+        <ViewOptionsPopover markdownUrl={markdownUrl} githubUrl={githubUrl} />
+      </div>
+      <div
+        role="separator"
+        style={{
+          height: '1px',
+          background: 'currentColor',
+          opacity: 0.15,
+          marginTop: '8px',
+          marginBottom: '24px',
+        }}
+      />
+      <DocsBody>
+        <MDX
+          components={getMDXComponents({
+            a: createRelativeLink(source, page),
+          })}
+        />
+      </DocsBody>
+    </DocsPage>
+  );
+}
+
+export async function generateStaticParams() {
+  return source.generateParams();
+}
+
+export async function generateMetadata(props: PageProps<'/docs/[[...slug]]'>): Promise<Metadata> {
+  const params = await props.params;
+  const page = source.getPage(params.slug);
+  if (!page) notFound();
+
+  return {
+    title: page.data.title,
+    description: page.data.description,
+    openGraph: {
+      images: getPageImage(page).url,
+    },
+  };
+}
@@ -0,0 +1,11 @@
+import { source } from '@/lib/source';
+import { DocsLayout } from 'fumadocs-ui/layouts/docs';
+import { baseOptions } from '@/lib/layout.shared';
+
+export default function Layout({ children }: LayoutProps<'/docs'>) {
+  return (
+    <DocsLayout tree={source.pageTree} {...baseOptions()}>
+      {children}
+    </DocsLayout>
+  );
+}
@@ -0,0 +1,14 @@
+@import 'tailwindcss';
+@import 'fumadocs-ui/css/neutral.css';
+@import 'fumadocs-ui/css/preset.css';
+@import 'fumadocs-openapi/css/preset.css';
+
+:root {
+  --color-fd-primary: hsl(43, 50%, 50%);
+  --color-fd-primary-foreground: hsl(222.2, 47.4%, 11.2%);
+}
+
+.dark {
+  --color-fd-primary: hsl(43, 50%, 45%);
+  --color-fd-primary-foreground: hsl(0, 0%, 95%);
+}
@@ -0,0 +1,17 @@
+import { RootProvider } from 'fumadocs-ui/provider/next';
+import './global.css';
+import { Inter } from 'next/font/google';
+
+const inter = Inter({
+  subsets: ['latin'],
+});
+
+export default function Layout({ children }: LayoutProps<'/'>) {
+  return (
+    <html lang="en" className={inter.className} suppressHydrationWarning>
+      <body className="flex flex-col min-h-screen">
+        <RootProvider>{children}</RootProvider>
+      </body>
+    </html>
+  );
+}
@@ -0,0 +1,10 @@
+import { getLLMText, source } from '@/lib/source';
+
+export const revalidate = false;
+
+export async function GET() {
+  const scan = source.getPages().map(getLLMText);
+  const scanned = await Promise.all(scan);
+
+  return new Response(scanned.join('\n\n'));
+}
@@ -0,0 +1,20 @@
+import { notFound } from 'next/navigation';
+import { getLLMText, source } from '@/lib/source';
+
+export const revalidate = false;
+
+export async function GET(_req: Request, { params }: RouteContext<'/llms.mdx/docs/[[...slug]]'>) {
+  const { slug } = await params;
+  const page = source.getPage(slug);
+  if (!page) notFound();
+
+  return new Response(await getLLMText(page), {
+    headers: {
+      'Content-Type': 'text/markdown',
+    },
+  });
+}
+
+export function generateStaticParams() {
+  return source.generateParams();
+}
@@ -0,0 +1,36 @@
+import { getPageImage, source } from '@/lib/source';
+import { notFound } from 'next/navigation';
+import { ImageResponse } from 'next/og';
+import { generate as DefaultImage } from 'fumadocs-ui/og';
+
+export const revalidate = false;
+
+export async function GET(
+  _req: Request,
+  { params }: RouteContext<'/og/docs/[...slug]'>,
+) {
+  const { slug } = await params;
+  const page = source.getPage(slug.slice(0, -1));
+  if (!page) notFound();
+
+  return new ImageResponse(
+    (
+      <DefaultImage
+        title={page.data.title}
+        description={page.data.description}
+        site="My App"
+      />
+    ),
+    {
+      width: 1200,
+      height: 630,
+    },
+  );
+}
+
+export function generateStaticParams() {
+  return source.getPages().map((page) => ({
+    lang: page.locale,
+    slug: getPageImage(page).segments,
+  }));
+}
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
James Pine	192979a762	docs	2026-03-16 05:11:21 -07:00
James Pine	e16cc42d53	enable Edit on GitHub and last updated on all doc pages	2026-03-16 04:45:40 -07:00
James Pine	f10e965003	rewrite docs root page, add screenshot	2026-03-16 04:12:11 -07:00
James Pine	a8968d4081	rewrite docs introduction based on README content	2026-03-16 04:09:50 -07:00
James Pine	7c4afbe4df	expand sidebar groups by default, remove stale plans reference	2026-03-16 04:08:46 -07:00
James Pine	a180fcc56f	redirect root to /docs	2026-03-16 04:06:38 -07:00
James Pine	1860b8dc92	remove plans/ from docs site	2026-03-16 04:05:14 -07:00
James Pine	1597937535	Merge branch 'main' into better-docs # Conflicts: # backend/main.py # docs/content/docs/plans/ADDING_TTS_ENGINES.md # docs/content/docs/plans/CUDA_BACKEND_SWAP.md # docs/content/docs/plans/CUDA_BACKEND_SWAP_FINAL.md # docs/content/docs/plans/EXTERNAL_PROVIDERS.md # docs/content/docs/plans/MLX_AUDIO.md # docs/content/docs/plans/PROJECT_STATUS.md	2026-03-16 04:01:08 -07:00
Jamie PineandGitHub	ac41a89359	Merge pull request #285 from jamiepine/backend-refactor Backend refactor: modular architecture, style guide, tooling	2026-03-16 03:51:10 -07:00
James Pine	60c0fe3b92	isolate shutdown unload calls so one failure doesn't block the other	2026-03-16 03:50:28 -07:00
James Pine	c99828cf76	fix startup db session leak on error (rollback + close in finally)	2026-03-16 03:49:21 -07:00
James Pine	5c4b979480	suppress E402 for app.py (AMD env vars must precede torch import)	2026-03-16 03:47:56 -07:00
James Pine	2d1b0ae820	remove unused _get_cuda_dll_excludes function	2026-03-16 03:46:58 -07:00
James Pine	69486c2a77	handle null duration in story_items migration	2026-03-16 03:44:33 -07:00
James Pine	e9f63d6c57	reject model migration to subdirectory of source cache	2026-03-16 03:43:37 -07:00
James Pine	8906bee23e	fix docstring for find_voicebox_pid_on_port	2026-03-16 03:43:06 -07:00
James Pine	0dabb121c9	improve startup logging: version, platform, data dir, db stats Replace verbose startup messages with a clean summary: - App version, Python version, OS/arch - Database path (fix None display), data directory - Profile and generation counts - Backend, GPU, model cache path - Clean up stale loading_model status on startup - Remove noisy progress manager log line	2026-03-16 03:42:39 -07:00
James Pine	944ba227ca	soften select focus indicator opacity	2026-03-16 03:22:36 -07:00
James Pine	473bb3e9fb	fix take-label race in regeneration, add accessible focus to select - Use DB COUNT query instead of list length for take-N label to avoid TOCTOU race between list_versions and create_version - Add focus:bg-muted to SelectTrigger for keyboard focus visibility	2026-03-16 03:22:05 -07:00
James Pine	0d0b62ea93	address CodeRabbit review: fix 4 critical + 12 major issues Critical: - Remove dead backend.utils.validation PyInstaller hidden import - Fix story_items table rebuild to preserve track/trim/version columns - Guard cache migration against same source/destination path - Fix regeneration audio overwrite (use random uuid suffix per take) Major: - Engine selector: validate language on Qwen switch, clear stale modelSize - Sync language validation regex between profile create and generate (22 langs) - Guard CUDA download against duplicate concurrent requests - Only set model_size for engines that support multiple sizes - Fix 404 swallowed by generic except in history export - Validate audio_path before FileResponse in export-audio - Transcription: stream uploads in 1MB chunks, use robust cache check, call complete_download() on Whisper download success - Set clean version as default when effects chain validation fails - Return explicit error when Windows port occupied by non-voicebox process	2026-03-16 03:12:01 -07:00
James Pine	798cd40f05	delete stale planning docs	2026-03-16 02:59:24 -07:00
James Pine	3187344f01	add model loading status, effects preset dropdown, clean up UI Backend: - Generation service reports 'loading_model' status only when model is not yet in memory, then 'generating' once inference starts - Migrate hf_offline_patch.py from print() to logging module - Update ADDING_TTS_ENGINES.md for post-refactor file paths Frontend: - HistoryTable shows 'Loading model...' vs 'Generating...' based on step - FloatingGenerateBox: replace instruct toggle + inline effects editor with an effects preset dropdown (third dropdown after language and engine) - Instruct UI removed for now (form field preserved for future models) - Remove focus ring from Select component globally	2026-03-16 02:58:41 -07:00
James Pine	8efcc95606	update Cargo.lock	2026-03-16 02:20:19 -07:00
James Pine	87cab9473d	gitignore: stop tracking tauri/src-tauri/gen/Assets.car Compiled Xcode asset catalog gets regenerated every build. No reason to track it.	2026-03-16 02:20:02 -07:00
James Pine	7b0fbfb567	rewrite backend README, remove completed refactor plan, update style guide Replace the outdated backend README (473 lines of stale API docs and pre-refactor file tree) with a concise architecture document covering module structure, request flow, backend selection, API domain overview, and development commands. Delete REFACTOR_PLAN.md -- all phases are complete. Update STYLE_GUIDE.md to remove refactor plan references and replace the verbose target layout with the current actual structure.	2026-03-16 02:18:34 -07:00
James Pine	7c1ea0a1e1	fix: replace netstat with TcpStream + PowerShell for port detection (#277 ) On Windows, Voicebox shelled out to netstat.exe on startup to check for existing server processes. On systems with corrupted DLLs, netstat fails with 0xc0000142, causing an infinite loading loop. Replace with: - TcpStream::connect_timeout() for port-in-use checks (pure Rust) - PowerShell Get-NetTCPConnection for port-to-PID lookup (built-in cmdlet) - tasklist for process name verification (unchanged) Closes #277	2026-03-16 02:15:26 -07:00
James Pine	b3012ed10c	move CRUD and service modules into services/, platform_detect into utils/ Move 9 business-logic modules from the backend root into services/: channels, effects, history, profiles, stories, versions, export_import, transcribe, tts. Move platform_detect.py into utils/. Backend root now contains only infrastructure (app, main, config, server, models, build_binary) and docs. All 94 routes verified.	2026-03-16 02:15:20 -07:00
James Pine	88536d27f7	extract routes from main.py into domain routers (Phase 4) Split the 2,578-line main.py (90 routes) into 12 domain-specific router modules under routes/. main.py is now a 45-line entry point. New structure: - app.py: FastAPI instance, CORS, startup/shutdown, safe_content_disposition - routes/: health, profiles, channels, generations, history, transcription, stories, effects, audio, models, tasks, cuda - services/cuda.py: moved from cuda_download.py Also includes Phase 5 database/ package (from parallel agent): - database/__init__.py re-exports all symbols for backward compat - database/models.py, session.py, migrations.py, seed.py All 90 routes verified registered and app imports cleanly.	2026-03-16 02:03:15 -07:00
James Pine	89d6e364d4	move pyproject	2026-03-16 01:48:26 -07:00
James Pine	b7781951df	comment cleanup	2026-03-16 01:46:19 -07:00
James Pine	fe19a9ca47	add style guide, ruff config, generation service extraction, remove Makefile - Add backend/STYLE_GUIDE.md covering formatting, imports, types, docstrings, comments, error handling, async, logging, and naming conventions - Add pyproject.toml with ruff linter/formatter config (ERA, FIX, isort, pyupgrade) - Extract generation service (Phase 3): unified run_generation() replaces three duplicated closures, serial queue moved to services/task_queue.py - Delete Makefile in favor of justfile; update all references - Add Python lint/format/test commands to justfile (check-python, fix-python, test) - Install ruff, pytest, pytest-asyncio as dev tools in setup-python - Update REFACTOR_PLAN.md with Phase 3 and Phase 7 completion	2026-03-16 01:35:59 -07:00
Jamie Pine	439fedcbf2	update refactor plan with phase 1+2 progress	2026-03-16 01:10:59 -07:00
Jamie Pine	0813a3d9d6	refactor: remove dead code, deduplicate backends Phase 1 - delete dead code: - studio.py, migrate_add_instruct.py, utils/validation.py - duplicate _profile_to_response in main.py, duplicate asyncio import - pointless _get_profiles_dir/_get_generations_dir wrappers - duplicate LANGUAGE_CODE_TO_NAME and WHISPER_HF_REPOS constants Phase 2 - extract backends/base.py with shared utilities: - is_model_cached() replaces 7 copy-pasted HF cache checks - get_torch_device() replaces 5 device detection methods - combine_voice_prompts() replaces 5 identical implementations - model_load_progress() ctx manager replaces progress boilerplate in all backends - patch_chatterbox_f32() replaces identical monkey-patches in both chatterbox backends net -1078 lines across the backend	2026-03-16 01:10:02 -07:00
Jamie Pine	9514c6596c	migrations	2026-03-16 00:54:13 -07:00
Jamie Pine	4e84415da7	refactor start	2026-03-16 00:52:23 -07:00
Jamie Pine	82cd4bf2ef	Add dynamic download redirect routes and update README links	2026-03-15 23:22:03 -07:00
Jamie Pine	3c30c5bec1	Update README for v0.2.x: multi-engine, effects, 23 languages, fix download links	2026-03-15 17:12:47 -07:00
Jamie Pine	c9d7bc4f27	Fix macOS download links to use .dmg instead of .app.tar.gz	2026-03-15 17:03:17 -07:00
James Pine	34e17bd469	Fix LuxTTS + Chatterbox in prod: bundle espeak/perth data, fix multiprocessing - collect-all piper_phonemize to bundle espeak-ng-data for LuxTTS phonemization - Set ESPEAK_DATA_PATH in frozen builds so the C library finds bundled data - collect-all perth to bundle pretrained watermark model for Chatterbox - Add multiprocessing.freeze_support() to fix resource_tracker subprocess crash	2026-03-15 16:02:09 -07:00
James Pine	aada13a5c9	Collect all inflect files for PyInstaller (fixes typeguard inspect.getsource)	2026-03-15 14:32:10 -07:00
James Pine	de8558d197	Fix prod build: download progress, robust stderr, full tracebacks - Force tqdm disable=False in TrackedTqdm so byte progress works in prod (huggingface_hub disables tqdm based on logger level, which prevents self.n from updating — our progress tracking needs the counter even though we don't render to terminal) - Harden devnull redirect to test writability, not just None check - Add full traceback logging to all backend error handlers - Add chatterbox/luxtts/zipvoice hidden imports and metadata to spec	2026-03-15 14:23:11 -07:00
Jamie Pine	9d79ea367a	Only use --noconsole on Windows, macOS/Linux need stdout for Tauri logs	2026-03-15 12:07:35 -07:00
Jamie Pine	04316f7adc	Copy metadata for requests/transformers/huggingface-hub to fix PyInstaller metadata lookup	2026-03-15 11:35:05 -07:00
Jamie Pine	4e4361d350	Fix noconsole crash: redirect None stdout/stderr to devnull on Windows	2026-03-15 11:27:35 -07:00
Jamie Pine	e9a249587c	Collect all linacodec files for PyInstaller (fixes inspect.getsource in Vocos)	2026-03-15 11:06:13 -07:00
Jamie Pine	d8a9ed7d15	Enable updater artifacts with v1Compatible for tauri-action sig generation	2026-03-15 10:54:12 -07:00
Jamie Pine	3dbf1c200e	Revert "Bump version: 0.2.3 → 0.2.4" This reverts commit `40fcb8d917`.	2026-03-15 10:20:31 -07:00
Jamie Pine	40fcb8d917	Bump version: 0.2.3 → 0.2.4	2026-03-15 10:18:51 -07:00
Jamie Pine	ad64d1c3d9	Collect all zipvoice files for PyInstaller (fixes source code error)	2026-03-15 10:18:40 -07:00
Jamie Pine	f826e45250	Install chatterbox-tts in CI release workflow	2026-03-15 10:17:23 -07:00
Jamie Pine	3d53c06c5b	Bump version: 0.2.2 → 0.2.3	2026-03-15 10:08:56 -07:00
James Pine	9835b9f6d4	fix: prevent stale release data by removing Next.js fetch cache Replace next: { revalidate: 600 } with cache: 'no-store' on GitHub API fetches so new releases show up within 5 minutes (in-memory cache only, no Next.js/Vercel cache layer on top).	2026-03-15 10:07:50 -07:00
Jamie Pine	a15dd30b1e	Update tauri-action to v0.6 to fix updater JSON and signature generation	2026-03-15 10:05:36 -07:00
Jamie Pine	1d343ac071	Treat missing/draft releases as up-to-date instead of showing error	2026-03-15 09:52:17 -07:00
James Pine	ca602de0ae	fix: don't reset audio player when unmuting during playback	2026-03-15 09:29:44 -07:00
James Pine	cdc0293ca8	feat: add /linux-install page with build-from-source instructions Linux download card now links to /linux-install instead of a direct binary download. The page explains the CI situation and gives clone + setup + build commands.	2026-03-15 09:17:30 -07:00
Jamie Pine	e7f749f082	Add luxtts/zipvoice hidden imports to PyInstaller build	2026-03-15 09:13:59 -07:00
Jamie Pine	d42e926e5c	Bump version: 0.2.1 → 0.2.2	2026-03-15 09:02:10 -07:00
Jamie Pine	32768ea874	Add chatterbox hidden imports to PyInstaller build	2026-03-15 09:00:13 -07:00
James Pine	b585e18ccf	fix: fade in hero background glow to avoid Safari rendering flash	2026-03-15 08:53:08 -07:00
Jamie Pine	655910457f	Auto-update CUDA binary on app update: check version on startup, download if stale	2026-03-15 08:46:17 -07:00
James Pine	d6984f1057	fix: remove mix-blend-lighten and drop-shadow causing boxes in Safari	2026-03-15 08:45:40 -07:00
James Pine	a637aebe69	feat: show version and total download count on landing page Fetches download counts across all GitHub releases (paginated) and displays version, total downloads, and platform list below the CTA.	2026-03-15 08:37:23 -07:00
Jamie Pine	2e6efa00a2	Refactor documentation structure and dependencies for migration to Fumadocs - Updated `.gitignore` to include new build and generated content directories. - Removed outdated Mintlify configuration files and documentation. - Introduced new `MIGRATION.md` to outline the transition from Mintlify to Fumadocs. - Added `mdx-components.tsx` for MDX component configuration and compatibility. - Updated `package.json` and `next.config.mjs` for new dependencies and Next.js configuration. - Created `source.config.ts` for content source configuration. - Added OpenAPI specification in `openapi.json` for API documentation. - Removed legacy files and adjusted project structure to align with Fumadocs conventions.	2026-02-02 23:29:35 -08:00
Jamie Pine	788a04f265	Merge branch 'main' into better-docs	2026-02-02 23:18:06 -08:00
Jamie Pine	5cb54ee03c	Update API documentation and enhance server configuration - Added server configurations for local and production environments in `main.py`. - Removed outdated authentication and generation API documentation files. - Updated documentation structure to reflect the removal of deprecated API endpoints. - Adjusted links in the quick start and developer setup documentation to point to the new API reference. - Enhanced global CSS styles for improved theming support.	2026-01-31 01:45:42 -08:00
Jamie Pine	0922845101	disable cuda for 0.1.12	2026-01-31 01:44:34 -08:00
Jamie Pine	64dd29d35a	Add initial setup for Fumadocs documentation migration - Created new directory structure for documentation under `/docs2`. - Added `.gitignore` to exclude build artifacts and dependencies. - Introduced `package.json`, `next.config.mjs`, and `postcss.config.mjs` for project configuration. - Implemented MDX components in `mdx-components.tsx` for rendering documentation. - Migrated existing documentation content and created new files for auto-updater and other features. - Established compatibility layer for Mintlify components in `mintlify-compat.tsx`. - Set up OpenAPI documentation in `openapi.json`. - Updated README and migration guide to reflect new structure and usage instructions. - Ensured all components and pages are ready for development and deployment with Fumadocs.	2026-01-30 23:32:45 -08:00
				`@@ -0,0 +1 @@`
				`# Services layer — generation orchestration and background task management.`