Generations now return immediately with a 'generating' status and appear
in history right away. TTS runs in a serial background queue to avoid
GPU contention. Users can kick off multiple generations without blocking.
- Async POST /generate creates DB record immediately, queues TTS work
- Serial generation queue prevents concurrent GPU access (Metal/CUDA/CPU)
- SSE endpoint GET /generate/{id}/status for real-time completion tracking
- Retry endpoint POST /generate/{id}/retry for failed generations
- Store engine and model_size on generation records for retry support
- History cards show animated loader (react-loaders) for generating/playing
- Failed generations show retry button instead of actions menu
- Model downloads happen inline in the queue instead of rejecting with 202
- Stale 'generating' records marked as failed on server startup
- Autoplay on generate setting (default: on)
- Show engine name on generation cards
- Remove sidebar generation spinner
- Checkbox alignment fix in settings
- Add model folder migration with byte-level progress tracking (backend + UI)
- Custom models directory support via VOICEBOX_MODELS_DIR env var passed to sidecar
- Hardcoded model descriptions displayed in model detail cards
- Open model folder button in storage location row
- Remove 'not downloaded' badge from model cards
- Fix server settings scroll offset for audio player
- Fix shell open permission to allow file paths
- Add normalize toggle to generation settings
Text exceeding max_chunk_chars (default 800) is automatically split at
sentence boundaries, generated per-chunk, and concatenated with a 50ms
crossfade. Works with all engines (Qwen, LuxTTS, Chatterbox, Turbo).
- Abbreviation-aware sentence splitter (Dr., Mr., e.g., decimals)
- CJK sentence-ending punctuation support
- Paralinguistic tag preservation ([laugh], [cough], etc.)
- Per-chunk seed variation to avoid correlated RNG artefacts
- Per-chunk Chatterbox trim (catches hallucination at each boundary)
- max_chunk_chars exposed as per-request param on GenerationRequest
- Text max_length raised to 50,000 characters
Closes#99
- Initialize lastSerializedRef to empty string so first-mount hydration
always runs (fixes initial value not rendering)
- Guard arrow-key menu nav against empty filteredTags (avoids NaN index)
- Disable ARIA role/multiline and detach event handlers when disabled
- Add onBlur to close autocomplete dropdown when editor loses focus
- Chain exception with 'from e' in unload endpoint for better tracebacks
Apply local-var-then-assign pattern to chatterbox_backend.py (multilingual)
to match the turbo backend. Also use _current_model_size fallback in
unload, delete, and status endpoints for consistent Qwen model size checks.
The actual dtype mismatch was in S3Tokenizer.log_mel_spectrogram, not
VoiceEncoder.forward. librosa.load returns float64 numpy, which
torch.from_numpy preserves as double. The STFT output (double) then
hits _mel_filters (float32) in a matmul at s3tokenizer.py:163.
Now patching both entry points after model load:
1. S3Tokenizer.log_mel_spectrogram — cast audio to float32 before STFT
2. VoiceEncoder.forward — cast mels to float32 before LSTM
Remove debug traceback logging (no longer needed).
- POST /models/{model_name}/unload — unloads a specific model from
memory without deleting from disk, supports all engine types
- Frontend: Unload button in model detail dialog when model is loaded
- Delete button remains disabled while loaded (unload first)
The previous approach of patching librosa.load didn't work because
melspectrogram itself performs float64 math (numpy dot, signal.lfilter)
regardless of input dtype. The actual mismatch happens when pack()
creates a float64 tensor from the mel arrays and passes it into the
float32 LSTM weights in VoiceEncoder.forward().
Fix by monkey-patching VoiceEncoder.forward() to call mels.float()
before the LSTM, ensuring the input always matches the model dtype.
soundfile cannot infer format from .tmp extension, causing all
generations to fail with 'No format specified and unable to get
format from file extension'
The upstream VoiceEncoder's melspectrogram only casts to float32 when
hp.normalized_mels is True (it defaults to False), so librosa's float64
output flows through as double tensors into float32 model weights,
causing 'expected m1 and m2 to have the same dtype, but got: float !=
double'. Fix by monkey-patching prepare_conditionals in both Chatterbox
and Chatterbox Turbo backends to ensure librosa.load returns float32.
Both PyTorch and MLX backends silently dropped the language parameter —
it was accepted by generate() but never forwarded to the underlying
Qwen3-TTS model, causing it to default to auto-detection which
frequently confuses similar languages (e.g. Portuguese for Spanish).
- Add LANGUAGE_CODE_TO_NAME mapping (ISO 639-1 to full name) to both backends
- PyTorch: pass language= to generate_voice_clone()
- MLX: pass lang_code= to all 4 model.generate() call sites
- Frontend: auto-sync generation form language with selected voice profile
Closes#97
- save_audio() now writes to .tmp then os.replace() for atomic writes
- /generate endpoint catches OSError with specific messages for ENOENT, EACCES, ENOSPC, and BrokenPipeError
- New /health/filesystem endpoint checks directory existence, write permissions, and disk space
- New DirectoryCheck and FilesystemHealthResponse models
Cherry-picked and expanded from #178 (@Vaibhavee89)
Cherry-picked and adapted from PR #89 and #214:
- Linux audio capture via PulseAudio/PipeWire monitor sources (cpal)
- AMD ROCm GPU support: HSA_OVERRIDE_GFX_VERSION env var, ROCm detection
- Whisper Turbo model (openai/whisper-large-v3-turbo) in all endpoints
- Cleaner Whisper language handling via generate_kwargs
- tauri::async_runtime::spawn fix to prevent panic on app shutdown
- Enable Linux (ubuntu-22.04) in release CI matrix
- New ChatterboxTurboTTSBackend wrapping ChatterboxTurboTTS (ResembleAI/chatterbox-turbo)
- English-only 350M model with paralinguistic tag support ([laugh], [cough], [chuckle])
- Bypasses upstream token=True bug by calling snapshot_download(token=None) + from_local()
- Same CPU-on-macOS forcing and torch.load monkey-patching as multilingual backend
- Full engine integration: generate, stream, model status/download/delete endpoints
- Language dropdown now shows only languages supported by the selected engine
- Per-engine language maps: Qwen (10), LuxTTS (en), Chatterbox (23), Turbo (en)
- Auto-switches to English when selecting English-only engines
- Backend language regex expanded to accept all 23 Chatterbox languages
- Add HFProgressTracker to LuxTTS and Chatterbox backends so tqdm-based
file-level download progress reaches the frontend (previously only Qwen
had this, LuxTTS/Chatterbox showed a static spinner)
- Add progress/current/total/filename fields to ActiveDownloadTask so the
/tasks/active polling endpoint carries progress data
- Show inline progress bar + bytes in the model list and detail modal,
poll at 1s during active downloads (5s otherwise)
- Fix GpuAcceleration crash: cudaStatusLoading was referenced before
initialization in its own useQuery declaration
chatterbox-tts 0.1.6 pins numpy<1.26 and torch==2.6 which are
incompatible with Python 3.12+. Install with --no-deps and list
its sub-dependencies explicitly in requirements.txt.
Also removes HFProgressTracker from chatterbox backend to avoid
'generator didn't stop after throw()' errors from tqdm patching.
- New ChatterboxTTSBackend wrapping ChatterboxMultilingualTTS (ResembleAI/chatterbox)
- Supports 23 languages including Hebrew, forces CPU on macOS (MPS issue)
- Monkey-patches torch.load for CPU loading, forces eager attention for compatibility
- trim_tts_output utility cuts trailing silence/hallucination from Chatterbox output
- Full engine integration: /generate, /generate/stream, model status/download/delete
- Hebrew (he) added to supported languages in frontend and backend validation
- Single flat model dropdown extended with Chatterbox option in both generation UIs
- ModelManagement UI groups LuxTTS and Chatterbox under 'Other Voice Models' section
- Add threading lock to get_tts_backend_for_engine() to prevent race
condition where concurrent requests could create duplicate backend
instances (double-checked locking pattern)
- Fix LuxTTS generate: call .detach().cpu() before .numpy() so it
works on GPU/MPS devices, not just CPU
- Store background download tasks in a module-level set to prevent
garbage collection before completion (asyncio.create_task fire-and-
forget pattern)
- Deduplicate cache_key computation in LuxTTS create_voice_prompt
- Prefix unused sr variable with underscore
- Combine engine + model size into one flat dropdown (Qwen3-TTS 1.7B,
Qwen3-TTS 0.6B, LuxTTS) in both FloatingGenerateBox and GenerationForm
- Add linacodec git dep to requirements.txt (uv-only source, pip can't
resolve it from Zipvoice's pyproject.toml)
- Remove redundant transitive deps from requirements.txt
- Quiet the sidecar setup script (was printing misleading instructions)
piper-phonemize has no PyPI wheels — needs custom find-links URL
from k2-fsa.github.io. Removed redundant transitive deps that
Zipvoice already declares.
Introduce LuxTTS (ZipVoice) alongside Qwen TTS, enabling users to choose
between engines at generation time. LuxTTS offers fast, English-focused
voice cloning at 48kHz with ~1GB VRAM.
Backend:
- Add LuxTTSBackend with encode_prompt/generate_speech integration
- Multi-engine registry (get_tts_backend_for_engine) replacing singleton
- Engine-prefixed voice prompt cache keys to avoid collisions
- Engine field on GenerationRequest (default 'qwen' for backward compat)
- Engine dispatch in /generate and /generate/stream endpoints
- LuxTTS in model status, download, and delete maps
Frontend:
- TTS Engine selector dropdown in GenerationForm (Qwen TTS / LuxTTS)
- Conditionally hide Model Size and Delivery Instructions for LuxTTS
- Engine field added to TypeScript types and Zod schema
- LuxTTS section in Model Management page
- Use YAML block scalar for inline run with colons (build-cuda.yml)
- Explicitly set VOICEBOX_BACKEND_VARIANT=cpu instead of setdefault (server.py)
- Use Path.replace() for atomic move on all platforms (cuda_download.py)
- Log actual exception in checksum fetch warning (cuda_download.py)
Add the ability to download a CUDA-enabled backend binary (~2.4 GB) and
swap it in via a backend-only restart, solving the #1 user pain point
(19 open 'GPU not detected' issues caused by GitHub's 2 GB asset limit).
Backend:
- cuda_download.py: download from R2 (primary) or GitHub split-parts
(fallback), SHA-256 verification, atomic writes, progress via SSE
- 4 new endpoints: GET/POST/DELETE /backend/cuda-*, GET cuda-progress
- server.py: --version flag, auto-detect variant from binary name
- build_binary.py: --cuda flag for CUDA PyInstaller builds
- split_binary.py: split large binaries into <2GB GitHub Release assets
- CI workflow for building CUDA binary
Tauri:
- restart_server command (stop -> wait -> start)
- start_server prefers CUDA binary from {data_dir}/backends/ if present
- Version mismatch check: runs --version before launching CUDA binary
Frontend:
- GpuAcceleration component: download, progress, restart, switch, delete
- API client + types for CUDA status and management
- Platform lifecycle: restartServer() on Tauri/Web
- Aggressive 1s health polling during restart for fast reconnection
- Fix transcribe_audio to use whisper-large-v3 mapping (not openai/whisper-large)
- Propagate error field in progress-only fallback path for get_active_tasks
- Use removed return value in cancel endpoint to vary response message
- Add error rollback to handleCancel with toast on failure
- Make isCancelling per-model instead of global
- Fix inverted chevron icons in Problems panel
- Move all clears under lock in clear_all_tasks
- Simplify cancel_download to use dict.pop()
- Add cancel (X) button on downloading and errored model items
- Add collapsible Problems panel (VS Code-style) showing error details
- Add "Clear All" button to reset all stale download/error state
- Add POST /models/download/cancel endpoint to dismiss individual downloads
- Add POST /tasks/clear endpoint to reset all task and progress state
- Include error messages in /tasks/active response for visibility
- Capture SSE error messages client-side for immediate display
- Fix whisper-large using wrong HF repo (openai/whisper-large → openai/whisper-large-v3)
- Fix Whisper HF repo mapping in both PyTorch and MLX backends
- Shorten error toast to point users to Problems panel instead of wall of text
- Add validation in create_profile() to check for existing names before insert
- Add validation in update_profile() to prevent renaming to duplicate names
- Improve error handling in API endpoints with user-friendly messages
- Add comprehensive test suite for duplicate name validation
- Update CHANGELOG.md with fix details
This fix prevents database constraint violations and provides clear
error messages when users attempt to create or update profiles with
names that already exist in the database.
Implements offline mode patch for API stability issues:
- Add hf_offline_patch.py to monkey-patch huggingface_hub
- Force cache-only lookups before mlx_audio imports
- Create symlink from original Qwen repo to MLX community version
when only MLX version is cached
This fixes:
- Issue #150: Internet required even with cached models
- Issue #151: API crashes when HF network fails
The patch ensures that if models are locally cached, no network
requests are made to HuggingFace during speech generation.
The /generate endpoint created the voice prompt before loading the
user's requested model size. Since create_voice_prompt() internally
calls load_model_async(None), it fell back to the hardcoded default
of "1.7B", causing the 1.7B model to be downloaded even when the
user explicitly selected 0.6B.
This reorders the operations so the requested model is loaded first,
ensuring create_voice_prompt() and generate() use the correct model.
Co-Authored-By: Claude Opus 4.6 <[email protected]>
The distributed macOS aarch64 binary shipped without MLX acceleration despite
the model and backend code supporting it. Two root causes:
1. **OSError not caught in platform_detect.py**
PyInstaller bundles isolate the filesystem, so when MLX tries to load its
Metal shader libraries (.metallib) it raises OSError, not ImportError.
platform_detect.get_backend_type() only caught ImportError, causing a
silent fallback to PyTorch even on Apple Silicon hardware.
Fix: broaden the except clause to (ImportError, OSError, RuntimeError)
and import mlx.core instead of mlx (forces native lib loading eagerly).
2. **collect_data_files used instead of collect_all for MLX**
build_binary.py and voicebox-server.spec used --collect-data /
collect_data_files for mlx and mlx_audio. This copies Python source and
pure-Python data, but NOT native shared libraries (.dylib, .metallib).
Fix: switch to --collect-all / collect_all which captures binaries too,
then pass them to Analysis(binaries=...) in the spec.
Result: macOS Apple Silicon users now get MLX inference (~4-5x faster than
PyTorch CPU), matching the performance documented in the README.
20 tests covering:
- All 6 default local origins are allowed
- Arbitrary external origins are blocked
- Preflight (OPTIONS) requests respect the allowlist
- VOICEBOX_CORS_ORIGINS env var extends the allowlist
- Edge cases: empty env, whitespace trimming, trailing commas
Tests use a minimal FastAPI app mirroring the real CORS config,
so they run without ML dependencies (torch, numpy, etc.).
The wildcard `allow_origins=["*"]` allows any website the user visits to
make requests to the local voicebox backend, potentially triggering TTS
generation or reading voice profiles without consent.
Restrict to the known Tauri webview and Vite dev server origins by
default. Users running in remote server mode can set
VOICEBOX_CORS_ORIGINS to allow additional origins.
Jamie Pine