fix(backend): pin miniaudio in requirements-mlx.txt (#505)

mlx-audio's STT path imports miniaudio, but we install mlx-audio
--no-deps to dodge its transformers>=5.x pin. Nothing else pulls
miniaudio transitively, so fresh Apple Silicon installs fail to
transcribe with ModuleNotFoundError: miniaudio. Listed explicitly
and updated the stale comments in requirements-mlx.txt and
release.yml that claimed it came from other engines.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

.github/workflows/release.yml

@@ -26,12 +26,45 @@ jobs:
             args: ""
             python-version: "3.12"
             backend: "pytorch"
+          - platform: "ubuntu-22.04"
+            # --config override disables updater-artifact generation on Linux.
+            # tauri.conf.json has createUpdaterArtifacts: "v1Compatible" which
+            # on Linux wants to synthesize a .AppImage.tar.gz by downloading
+            # linuxdeploy at build time — this is what silently hangs CI
+            # (see v0.4.2 round 2, 25 min of no output after rpm bundling).
+            # We ship deb+rpm only; Linux users update via apt/dnf, not the
+            # Tauri in-app updater.
+            args: '--target x86_64-unknown-linux-gnu --bundles deb,rpm --verbose --config {"bundle":{"createUpdaterArtifacts":false}}'
+            python-version: "3.12"
+            backend: "pytorch"
 
     runs-on: ${{ matrix.platform }}
 
     steps:
       - uses: actions/checkout@v4
 
+      # Ubuntu runners ship with ~14 GB free; pip + PyInstaller + torch can
+      # peak well above that during the build. Reclaim ~25 GB by pruning
+      # preinstalled toolchains we don't use. This is what likely tripped
+      # the March 2026 Linux release attempts (see commit 103e98b
+      # "github runners suck") — not a code issue, a disk-pressure one.
+      - name: Free up disk space (ubuntu)
+        if: contains(matrix.platform, 'ubuntu') || contains(matrix.platform, 'namespace')
+        # Pinned to v1.3.1 (SHA) — this job runs with contents: write and
+        # handles signing secrets later, so we don't want a floating ref.
+        uses: jlumbroso/free-disk-space@54081f138730dfa15788a46383842cd2f914a1be
+        with:
+          tool-cache: false
+          android: true
+          dotnet: true
+          haskell: true
+          # large-packages: true would `apt-get remove '^llvm-.*'`, which
+          # cascade-removes reverse deps that won't be pulled back in by the
+          # `llvm-dev` install below. The other flags already free ~20 GB,
+          # enough for the Python + torch + PyInstaller build.
+          large-packages: false
+          swap-storage: true
+
       - name: Install dependencies (ubuntu only)
         if: contains(matrix.platform, 'ubuntu') || contains(matrix.platform, 'namespace')
         run: |
@@ -73,8 +106,10 @@ jobs:
           # fine on transformers 4.57.x in practice (verified in dev), so install
           # them --no-deps. mlx-audio's other runtime deps (huggingface_hub,
           # librosa, numpy, numba, pyloudnorm) are already in requirements.txt;
-          # the rest (sounddevice, miniaudio, protobuf, sentencepiece, pyyaml,
-          # jinja2) are pulled in by other engines.
+          # miniaudio is in requirements-mlx.txt (needed by mlx_audio.stt,
+          # not transitively pulled by anything else — see issue #505); the
+          # rest (sounddevice, protobuf, sentencepiece, pyyaml, jinja2) are
+          # pulled in by other engines.
           pip install --no-deps mlx-lm==0.31.1
           pip install --no-deps mlx-audio==0.4.1
 
@@ -133,6 +168,21 @@ jobs:
           p12-file-base64: ${{ secrets.APPLE_CERTIFICATE }}
           p12-password: ${{ secrets.APPLE_CERTIFICATE_PASSWORD }}
 
+      - name: Disk / environment snapshot (pre-bundle debug)
+        if: contains(matrix.platform, 'ubuntu') || contains(matrix.platform, 'namespace')
+        run: |
+          echo "=== df -h ==="
+          df -h
+          echo "=== free -h ==="
+          free -h
+          echo "=== Rust / Cargo ==="
+          rustc --version
+          cargo --version
+          echo "=== Bun ==="
+          bun --version
+          echo "=== Tauri CLI ==="
+          cd tauri && bun run tauri --version
+
       - name: Extract release notes from CHANGELOG.md
         id: changelog
         shell: bash
@@ -156,7 +206,13 @@ jobs:
             echo "CHANGELOG_EOF"
           } >> "$GITHUB_OUTPUT"
 
+      # Linux hang watchdog: previous releases silently wedged inside tauri
+      # bundling (possibly linuxdeploy/AppImage download, possibly cargo link).
+      # Cap the step at 30 min so we get logs instead of waiting out the 6hr
+      # job timeout. Other platforms historically complete in ~25 min, so 45
+      # is comfortable.
       - uses: tauri-apps/[email protected]
+        timeout-minutes: ${{ (contains(matrix.platform, 'ubuntu') || contains(matrix.platform, 'namespace')) && 30 || 45 }}
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           TAURI_SIGNING_PRIVATE_KEY: ${{ secrets.TAURI_SIGNING_PRIVATE_KEY }}
@@ -168,6 +224,9 @@ jobs:
           APPLE_PROVIDER_SHORT_NAME: ${{ secrets.APPLE_PROVIDER_SHORT_NAME }}
           APPLE_API_ISSUER: ${{ secrets.APPLE_API_ISSUER }}
           APPLE_API_KEY: ${{ secrets.APPLE_API_KEY }}
+          # Stream subprocess stdout/stderr so the hang is visible in logs.
+          CARGO_TERM_VERBOSE: "true"
+          RUST_BACKTRACE: "1"
         with:
           projectPath: tauri
           tagName: v__VERSION__

CONTRIBUTING.md

@@ -359,7 +359,7 @@ Releases are managed by maintainers:
 
 ## Troubleshooting
 
-See [docs/TROUBLESHOOTING.md](docs/TROUBLESHOOTING.md) for common issues and solutions.
+See [docs/content/docs/overview/troubleshooting.mdx](docs/content/docs/overview/troubleshooting.mdx) for common issues and solutions.
 
 **Quick fixes:**
 
@@ -372,7 +372,7 @@ See [docs/TROUBLESHOOTING.md](docs/TROUBLESHOOTING.md) for common issues and sol
 - Open an issue for bugs or feature requests
 - Check existing issues and discussions
 - Review the codebase to understand patterns
-- See [docs/TROUBLESHOOTING.md](docs/TROUBLESHOOTING.md) for common issues
+- See [docs/content/docs/overview/troubleshooting.mdx](docs/content/docs/overview/troubleshooting.mdx) for common issues
 
 ## Additional Resources
 

README.md

@@ -33,7 +33,8 @@
   <a href="https://docs.voicebox.sh">Docs</a> •
   <a href="#download">Download</a> •
   <a href="#features">Features</a> •
-  <a href="#api">API</a>
+  <a href="#api">API</a> •
+  <a href="docs/content/docs/overview/troubleshooting.mdx">Troubleshooting</a>
 </p>
 
 <br/>
@@ -91,6 +92,8 @@ Voicebox is a **local-first voice cloning studio** — a free and open-source al
 
 > **Linux** — Pre-built binaries are not yet available. See [voicebox.sh/linux-install](https://voicebox.sh/linux-install) for build-from-source instructions.
 
+> **Having trouble?** See the [Troubleshooting Guide](docs/content/docs/overview/troubleshooting.mdx) for common install, generation, model-download, and GPU issues.
+
 ---
 
 ## Features

app/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@voicebox/app",
-  "version": "0.4.1",
+  "version": "0.4.2",
   "private": true,
   "type": "module",
   "scripts": {

app/src/components/AppFrame/AppFrame.tsx

@@ -1,5 +1,6 @@
 import { useRouterState } from '@tanstack/react-router';
 import { TitleBarDragRegion } from '@/components/TitleBarDragRegion';
+import { AudioKeepAlive } from '@/components/AudioPlayer/AudioKeepAlive';
 import { AudioPlayer } from '@/components/AudioPlayer/AudioPlayer';
 import { StoryTrackEditor } from '@/components/StoriesTab/StoryTrackEditor';
 import { TOP_SAFE_AREA_PADDING } from '@/lib/constants/ui';
@@ -26,6 +27,7 @@ export function AppFrame({ children }: AppFrameProps) {
       className={cn('h-screen bg-background flex flex-col overflow-hidden', TOP_SAFE_AREA_PADDING)}
     >
       <TitleBarDragRegion />
+      <AudioKeepAlive />
       {children}
       {showTrackEditor ? (
         <StoryTrackEditor storyId={story.id} items={story.items} />

app/src/components/AudioPlayer/AudioKeepAlive.tsx

@@ -0,0 +1,85 @@
+import { useEffect, useRef } from 'react';
+import { debug } from '@/lib/utils/debug';
+
+// WKWebView tears down the app's CoreAudio output when idle for long enough,
+// and a JS-level reload (cmd+R) does NOT restore it — only relaunching the
+// Tauri app does. Keeping a silent <audio> element looping forever prevents
+// the OS audio session from ever going dormant.
+//
+// Real silence (zero PCM samples) at full volume is preferred over a muted
+// element: browsers/WebKit can optimize muted media away, which defeats the
+// purpose of holding the session open.
+
+function buildSilentWavUrl(seconds = 1, sampleRate = 8000): string {
+  const numSamples = seconds * sampleRate;
+  const bytes = 44 + numSamples * 2;
+  const buffer = new ArrayBuffer(bytes);
+  const view = new DataView(buffer);
+  const write = (offset: number, str: string) => {
+    for (let i = 0; i < str.length; i++) view.setUint8(offset + i, str.charCodeAt(i));
+  };
+  write(0, 'RIFF');
+  view.setUint32(4, bytes - 8, true);
+  write(8, 'WAVE');
+  write(12, 'fmt ');
+  view.setUint32(16, 16, true);
+  view.setUint16(20, 1, true);
+  view.setUint16(22, 1, true);
+  view.setUint32(24, sampleRate, true);
+  view.setUint32(28, sampleRate * 2, true);
+  view.setUint16(32, 2, true);
+  view.setUint16(34, 16, true);
+  write(36, 'data');
+  view.setUint32(40, numSamples * 2, true);
+  return URL.createObjectURL(new Blob([buffer], { type: 'audio/wav' }));
+}
+
+export function AudioKeepAlive() {
+  const audioRef = useRef<HTMLAudioElement | null>(null);
+
+  useEffect(() => {
+    const url = buildSilentWavUrl(1, 8000);
+    const el = new Audio(url);
+    el.loop = true;
+    el.volume = 1;
+    el.preload = 'auto';
+    audioRef.current = el;
+
+    const tryPlay = () => {
+      if (!audioRef.current) return;
+      if (!audioRef.current.paused) return;
+      audioRef.current.play().catch((err) => {
+        debug.log('[AudioKeepAlive] play blocked (will retry on next gesture):', err);
+      });
+    };
+
+    tryPlay();
+
+    // Autoplay may be blocked until first user interaction — re-attempt then.
+    const onGesture = () => tryPlay();
+    window.addEventListener('pointerdown', onGesture, { once: false });
+    window.addEventListener('keydown', onGesture, { once: false });
+
+    // If the webview ever pauses the element on background, resume on return.
+    const onWake = () => {
+      if (!document.hidden) tryPlay();
+    };
+    document.addEventListener('visibilitychange', onWake);
+    window.addEventListener('focus', onWake);
+    window.addEventListener('pageshow', onWake);
+
+    return () => {
+      window.removeEventListener('pointerdown', onGesture);
+      window.removeEventListener('keydown', onGesture);
+      document.removeEventListener('visibilitychange', onWake);
+      window.removeEventListener('focus', onWake);
+      window.removeEventListener('pageshow', onWake);
+      el.pause();
+      el.src = '';
+      URL.revokeObjectURL(url);
+      audioRef.current = null;
+    };
+  }, []);
+
+  return null;
+}

backend/backends/__init__.py

@@ -177,7 +177,7 @@ def _get_qwen_model_configs() -> list[ModelConfig]:
     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
+        repo_0_6b = "mlx-community/Qwen3-TTS-12Hz-0.6B-Base-bf16"
     else:
         repo_1_7b = "Qwen/Qwen3-TTS-12Hz-1.7B-Base"
         repo_0_6b = "Qwen/Qwen3-TTS-12Hz-0.6B-Base"

backend/backends/mlx_backend.py

@@ -45,11 +45,9 @@ class MLXTTSBackend:
         Returns:
             HuggingFace Hub model ID for MLX
         """
-        # MLX model mapping
         mlx_model_map = {
             "1.7B": "mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16",
-            # 0.6B not yet converted to MLX format
-            "0.6B": "mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16",  # Fallback to 1.7B
+            "0.6B": "mlx-community/Qwen3-TTS-12Hz-0.6B-Base-bf16",
         }
 
         if model_size not in mlx_model_map:
@@ -195,6 +193,8 @@ class MLXTTSBackend:
 
         logger.info("Generating audio for text: %s", text)
 
+        model_name = f"qwen-tts-{self._current_model_size}"
+
         def _generate_sync():
             """Run synchronous generation in thread pool."""
             # MLX generate() returns a generator yielding GenerationResult objects
@@ -220,36 +220,40 @@ class MLXTTSBackend:
                 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:
-                # Try with voice cloning parameters if supported
-                if ref_audio:
-                    # Check if generate accepts ref_audio parameter
-                    import inspect
+            # Model is loaded → weights are on disk. Force offline so
+            # lazy tokenizer/config lookups inside mlx_audio don't hang
+            # when the user is disconnected (issue #462).
+            with force_offline_if_cached(True, model_name):
+                # Check if model supports voice cloning via generate method
+                # MLX API may support ref_audio parameter directly
+                try:
+                    # Try with voice cloning parameters if supported
+                    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
-                        for result in self.model.generate(text, ref_audio=ref_audio, ref_text=ref_text, lang_code=lang):
-                            audio_chunks.append(np.array(result.audio))
-                            sample_rate = result.sample_rate
+                        sig = inspect.signature(self.model.generate)
+                        if "ref_audio" in sig.parameters:
+                            # Generate with voice cloning
+                            for result in self.model.generate(text, ref_audio=ref_audio, ref_text=ref_text, lang_code=lang):
+                                audio_chunks.append(np.array(result.audio))
+                                sample_rate = result.sample_rate
+                        else:
+                            # Fallback: generate without voice cloning
+                            for result in self.model.generate(text, lang_code=lang):
+                                audio_chunks.append(np.array(result.audio))
+                                sample_rate = result.sample_rate
                     else:
-                        # Fallback: generate without voice cloning
+                        # No voice prompt, generate normally
                         for result in self.model.generate(text, lang_code=lang):
                             audio_chunks.append(np.array(result.audio))
                             sample_rate = result.sample_rate
-                else:
-                    # No voice prompt, generate normally
+                except Exception as e:
+                    # If voice cloning fails, try without it
+                    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
-            except Exception as e:
-                # If voice cloning fails, try without it
-                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:
@@ -343,6 +347,8 @@ class MLXSTTBackend:
         """
         await self.load_model_async(model_size)
 
+        progress_model_name = f"whisper-{self.model_size}"
+
         def _transcribe_sync():
             """Run synchronous transcription in thread pool."""
             # MLX Whisper transcription using generate method
@@ -351,7 +357,11 @@ class MLXSTTBackend:
             if language:
                 decode_options["language"] = language
 
-            result = self.model.generate(str(audio_path), **decode_options)
+            # Model is loaded → weights are on disk. Force offline so
+            # lazy tokenizer/config lookups don't hang when the user is
+            # disconnected (issue #462).
+            with force_offline_if_cached(True, progress_model_name):
+                result = self.model.generate(str(audio_path), **decode_options)
 
             # Extract text from result
             if isinstance(result, str):

backend/backends/pytorch_backend.py

@@ -172,13 +172,19 @@ class PyTorchTTSBackend:
                     # This shouldn't happen in practice, but handle it
                     return {"prompt": cached_prompt}, True
 
+        model_name = f"qwen-tts-{self._current_model_size}"
+
         def _create_prompt_sync():
             """Run synchronous voice prompt creation in thread pool."""
-            return self.model.create_voice_clone_prompt(
-                ref_audio=str(audio_path),
-                ref_text=reference_text,
-                x_vector_only_mode=False,
-            )
+            # Model is loaded → weights are on disk. Force offline so
+            # lazy tokenizer/config lookups inside qwen_tts don't hang
+            # when the user is disconnected (issue #462).
+            with force_offline_if_cached(True, model_name):
+                return self.model.create_voice_clone_prompt(
+                    ref_audio=str(audio_path),
+                    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)
@@ -221,19 +227,24 @@ class PyTorchTTSBackend:
         # Load model
         await self.load_model_async(None)
 
+        model_name = f"qwen-tts-{self._current_model_size}"
+
         def _generate_sync():
             """Run synchronous generation in thread pool."""
             # Set seed if provided
             if seed is not None:
                 manual_seed(seed, self.device)
 
-            # Generate audio - this is the blocking operation
-            wavs, sample_rate = self.model.generate_voice_clone(
-                text=text,
-                voice_clone_prompt=voice_prompt,
-                language=LANGUAGE_CODE_TO_NAME.get(language, "auto"),
-                instruct=instruct,
-            )
+            # Model is loaded → weights are on disk. Force offline so
+            # lazy tokenizer/config lookups inside qwen_tts don't hang
+            # when the user is disconnected (issue #462).
+            with force_offline_if_cached(True, model_name):
+                wavs, sample_rate = self.model.generate_voice_clone(
+                    text=text,
+                    voice_clone_prompt=voice_prompt,
+                    language=LANGUAGE_CODE_TO_NAME.get(language, "auto"),
+                    instruct=instruct,
+                )
             return wavs[0], sample_rate
 
         # Run blocking inference in thread pool to avoid blocking event loop
@@ -331,40 +342,46 @@ class PyTorchSTTBackend:
         """
         await self.load_model_async(model_size)
 
+        progress_model_name = f"whisper-{self.model_size}"
+
         def _transcribe_sync():
             """Run synchronous transcription in thread pool."""
             # Load audio
-            audio, sr = load_audio(audio_path, sample_rate=16000)
+            audio, _sr = load_audio(audio_path, sample_rate=16000)
 
-            # Process audio
-            inputs = self.processor(
-                audio,
-                sampling_rate=16000,
-                return_tensors="pt",
-            )
-            inputs = inputs.to(self.device)
-
-            # Generate transcription
-            # If language is provided, force it; otherwise let Whisper auto-detect
-            generate_kwargs = {}
-            if language:
-                forced_decoder_ids = self.processor.get_decoder_prompt_ids(
-                    language=language,
-                    task="transcribe",
+            # Model is loaded → weights are on disk. Force offline so
+            # `get_decoder_prompt_ids` and any lazy tokenizer lookups
+            # don't hang when the user is disconnected (issue #462).
+            with force_offline_if_cached(True, progress_model_name):
+                # Process audio
+                inputs = self.processor(
+                    audio,
+                    sampling_rate=16000,
+                    return_tensors="pt",
                 )
-                generate_kwargs["forced_decoder_ids"] = forced_decoder_ids
+                inputs = inputs.to(self.device)
 
-            with torch.no_grad():
-                predicted_ids = self.model.generate(
-                    inputs["input_features"],
-                    **generate_kwargs,
-                )
+                # Generate transcription
+                # If language is provided, force it; otherwise let Whisper auto-detect
+                generate_kwargs = {}
+                if language:
+                    forced_decoder_ids = self.processor.get_decoder_prompt_ids(
+                        language=language,
+                        task="transcribe",
+                    )
+                    generate_kwargs["forced_decoder_ids"] = forced_decoder_ids
 
-            # Decode
-            transcription = self.processor.batch_decode(
-                predicted_ids,
-                skip_special_tokens=True,
-            )[0]
+                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()
 

backend/backends/qwen_custom_voice_backend.py

@@ -28,6 +28,7 @@ from .base import (
     combine_voice_prompts as _combine_voice_prompts,
     model_load_progress,
 )
+from ..utils.hf_offline_patch import force_offline_if_cached
 
 logger = logging.getLogger(__name__)
 
@@ -104,18 +105,19 @@ class QwenCustomVoiceBackend:
             model_path = self._get_model_path(model_size)
             logger.info("Loading Qwen CustomVoice %s on %s...", model_size, self.device)
 
-            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,
-                )
+            with force_offline_if_cached(is_cached, model_name):
+                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,
+                    )
 
         self._current_model_size = model_size
         self.model_size = model_size
@@ -184,6 +186,7 @@ class QwenCustomVoiceBackend:
         await self.load_model_async(None)
 
         speaker = voice_prompt.get("preset_voice_id") or QWEN_CV_DEFAULT_SPEAKER
+        model_name = f"qwen-custom-voice-{self._current_model_size}"
 
         def _generate_sync():
             if seed is not None:
@@ -203,7 +206,11 @@ class QwenCustomVoiceBackend:
             if instruct:
                 kwargs["instruct"] = instruct
 
-            wavs, sample_rate = self.model.generate_custom_voice(**kwargs)
+            # Model is loaded → weights are on disk. Force offline so
+            # lazy tokenizer/config lookups inside qwen_tts don't hang
+            # when the user is disconnected (issue #462).
+            with force_offline_if_cached(True, model_name):
+                wavs, sample_rate = self.model.generate_custom_voice(**kwargs)
             return wavs[0], sample_rate
 
         audio, sample_rate = await asyncio.to_thread(_generate_sync)

backend/requirements-mlx.txt

@@ -3,6 +3,12 @@
 
 mlx>=0.30.0
 
+# miniaudio is a runtime dep of mlx-audio's STT path (mlx_audio.stt).
+# mlx-audio itself is installed --no-deps (see comment below), so we
+# must list miniaudio explicitly here or transcription fails on fresh
+# M1 installs with `ModuleNotFoundError: miniaudio` (issue #505).
+miniaudio>=1.59
+
 # NOTE: mlx-audio is intentionally not listed here. From 0.3.1 onward it
 # declares `transformers==5.0.0rc3` / `>=5.0.0`, which conflicts with the
 # `transformers<=4.57.6` cap in requirements.txt and breaks CI's clean
@@ -10,6 +16,7 @@ mlx>=0.30.0
 # mlx_audio.stt.load) works fine on transformers 4.57.x in practice.
 #
 # Install it via `pip install --no-deps mlx-audio==0.4.1` after this file
-# (see .github/workflows/release.yml). All other mlx-audio runtime deps
-# (huggingface_hub, librosa, miniaudio, mlx-lm, numba, numpy, protobuf,
-# pyloudnorm, sounddevice, tqdm) are already in requirements.txt.
+# (see .github/workflows/release.yml). Most other mlx-audio runtime deps
+# (huggingface_hub, librosa, mlx-lm, numba, numpy, protobuf, pyloudnorm,
+# sounddevice, tqdm) are already in requirements.txt or pulled in by
+# other engines.

backend/tests/test_audio_preprocess.py

@@ -0,0 +1,112 @@
+"""
+Unit tests for reference-audio preprocessing.
+
+Covers :func:`backend.utils.audio.preprocess_reference_audio` and
+:func:`backend.utils.audio.validate_and_load_reference_audio`.
+"""
+
+import sys
+from pathlib import Path
+
+import numpy as np
+import pytest
+import soundfile as sf
+
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+from utils.audio import (  # noqa: E402
+    preprocess_reference_audio,
+    validate_and_load_reference_audio,
+)
+
+
+SR = 24000
+
+
+def _tone(duration_s: float, amp: float = 0.3, freq: float = 220.0) -> np.ndarray:
+    n = int(duration_s * SR)
+    t = np.arange(n, dtype=np.float32) / SR
+    return (amp * np.sin(2 * np.pi * freq * t)).astype(np.float32)
+
+
+def test_peak_cap_scales_hot_input():
+    audio = _tone(3.0, amp=0.99)
+    out = preprocess_reference_audio(audio, SR)
+    assert np.abs(out).max() <= 0.951
+
+
+def test_peak_cap_leaves_moderate_input_untouched():
+    audio = _tone(3.0, amp=0.5)
+    out = preprocess_reference_audio(audio, SR)
+    assert np.isclose(np.abs(out).max(), 0.5, atol=1e-3)
+
+
+def test_dc_offset_removed():
+    audio = _tone(3.0, amp=0.3) + 0.1
+    out = preprocess_reference_audio(audio, SR)
+    assert abs(float(np.mean(out))) < 1e-3
+
+
+def test_silence_is_trimmed_with_padding_kept():
+    silence = np.zeros(int(SR * 1.0), dtype=np.float32)
+    speech = _tone(3.0, amp=0.3)
+    audio = np.concatenate([silence, speech, silence])
+    out = preprocess_reference_audio(audio, SR)
+    # Most of the 2s of leading/trailing silence should be gone, but the
+    # 3s of speech plus ~200ms of padding should remain.
+    assert len(audio) - len(out) >= SR, "expected >=1s of silence trimmed"
+    assert len(out) >= int(3.0 * SR), "speech body should be preserved"
+
+
+def test_clean_audio_is_not_padded_past_original_length():
+    # Well-recorded audio with no edge silence shouldn't get longer after
+    # preprocessing — otherwise a 29.9 s upload could be pushed past the
+    # 30 s max_duration ceiling downstream.
+    audio = _tone(3.0, amp=0.3)
+    out = preprocess_reference_audio(audio, SR)
+    assert len(out) <= len(audio)
+
+
+def test_empty_input_returns_empty():
+    out = preprocess_reference_audio(np.zeros(0, dtype=np.float32), SR)
+    assert out.size == 0
+
+
+def test_validate_accepts_previously_rejected_hot_file(tmp_path):
+    audio = _tone(3.0, amp=0.995)
+    path = tmp_path / "hot.wav"
+    sf.write(str(path), audio, SR)
+
+    ok, err, out_audio, out_sr = validate_and_load_reference_audio(str(path))
+
+    assert ok, f"expected pass, got error: {err}"
+    assert out_audio is not None
+    assert out_sr == SR
+    assert np.abs(out_audio).max() <= 0.951
+
+
+def test_validate_still_rejects_silent_input(tmp_path):
+    audio = np.zeros(int(SR * 3.0), dtype=np.float32)
+    path = tmp_path / "silent.wav"
+    sf.write(str(path), audio, SR)
+
+    ok, err, _, _ = validate_and_load_reference_audio(str(path))
+
+    assert not ok
+    assert err is not None
+    assert "too short" in err.lower() or "quiet" in err.lower()
+
+
+def test_validate_rejects_too_short(tmp_path):
+    audio = _tone(0.5, amp=0.3)
+    path = tmp_path / "short.wav"
+    sf.write(str(path), audio, SR)
+
+    ok, err, _, _ = validate_and_load_reference_audio(str(path))
+
+    assert not ok
+    assert "too short" in (err or "").lower()
+
+
+if __name__ == "__main__":
+    pytest.main([__file__, "-v"])

backend/tests/test_offline_guard.py

@@ -0,0 +1,118 @@
+"""
+Unit tests for the ``force_offline_if_cached`` helper.
+
+Verifies that the helper mutates the cached module constants in
+``huggingface_hub.constants`` and ``transformers.utils.hub`` — not just
+``os.environ`` — and that concurrent users are refcount-coordinated so
+one thread's exit can't strip another thread's offline protection.
+
+NOTE: These tests mutate process-global state in ``huggingface_hub.constants``
+and ``transformers.utils.hub``. They are not safe under cross-process
+parallelism (e.g. ``pytest-xdist`` with ``--dist=loadfile``/``loadscope``);
+run this file serially.
+"""
+
+import os
+import sys
+import threading
+from pathlib import Path
+
+import pytest
+
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+from utils.hf_offline_patch import force_offline_if_cached  # noqa: E402
+
+
+def _hf_const():
+    import huggingface_hub.constants as hf_const
+
+    return hf_const
+
+
+def _tf_hub():
+    import transformers.utils.hub as tf_hub
+
+    return tf_hub
+
+
+def test_mutates_cached_huggingface_hub_constant():
+    original = _hf_const().HF_HUB_OFFLINE
+    with force_offline_if_cached(True, "t"):
+        assert _hf_const().HF_HUB_OFFLINE is True
+    assert original == _hf_const().HF_HUB_OFFLINE
+
+
+def test_mutates_cached_transformers_constant():
+    original = _tf_hub()._is_offline_mode
+    with force_offline_if_cached(True, "t"):
+        assert _tf_hub()._is_offline_mode is True
+    assert original == _tf_hub()._is_offline_mode
+
+
+def test_sets_env_variable():
+    original = os.environ.get("HF_HUB_OFFLINE")
+    with force_offline_if_cached(True, "t"):
+        assert "1" == os.environ.get("HF_HUB_OFFLINE")
+    assert original == os.environ.get("HF_HUB_OFFLINE")
+
+
+def test_noop_when_not_cached():
+    before = _hf_const().HF_HUB_OFFLINE
+    with force_offline_if_cached(False, "t"):
+        assert before == _hf_const().HF_HUB_OFFLINE
+
+
+def test_nested_contexts_respect_refcount():
+    original = _hf_const().HF_HUB_OFFLINE
+    with force_offline_if_cached(True, "outer"):
+        assert _hf_const().HF_HUB_OFFLINE is True
+        with force_offline_if_cached(True, "inner"):
+            assert _hf_const().HF_HUB_OFFLINE is True
+        # inner exit must not restore while outer is still active
+        assert _hf_const().HF_HUB_OFFLINE is True
+    assert original == _hf_const().HF_HUB_OFFLINE
+
+
+def test_concurrent_threads_share_offline_window():
+    """A slow thread must keep seeing offline mode even if a peer exits first."""
+    original = _hf_const().HF_HUB_OFFLINE
+    observations: list[bool] = []
+    errors: list[Exception] = []
+    barrier = threading.Barrier(2)
+    fast_exited = threading.Event()
+
+    def slow():
+        try:
+            with force_offline_if_cached(True, "slow"):
+                barrier.wait(timeout=5)
+                assert fast_exited.wait(timeout=5), "fast thread did not exit"
+                observations.append(_hf_const().HF_HUB_OFFLINE)
+        except Exception as exc:  # noqa: BLE001
+            errors.append(exc)
+
+    def fast():
+        try:
+            with force_offline_if_cached(True, "fast"):
+                barrier.wait(timeout=5)
+        except Exception as exc:  # noqa: BLE001
+            errors.append(exc)
+        finally:
+            fast_exited.set()
+
+    t_slow = threading.Thread(target=slow)
+    t_fast = threading.Thread(target=fast)
+    t_slow.start()
+    t_fast.start()
+    t_slow.join(timeout=5)
+    t_fast.join(timeout=5)
+
+    assert not t_slow.is_alive(), "slow thread did not finish"
+    assert not t_fast.is_alive(), "fast thread did not finish"
+    assert not errors, errors
+    assert observations == [True], "slow thread lost offline protection"
+    assert original == _hf_const().HF_HUB_OFFLINE
+
+
+if __name__ == "__main__":
+    pytest.main([__file__, "-v"])

backend/tests/test_qwen_download.py

@@ -175,7 +175,7 @@ async def main():
     print("   ✅ Server is running")
 
     # Test model
-    model_name = "qwen-tts-0.6B"  # Note: 0.6B currently maps to 1.7B on MLX
+    model_name = "qwen-tts-0.6B"
     
     # Check current status
     print(f"\n📊 Checking status of {model_name}...")

backend/utils/audio.py

@@ -199,6 +199,66 @@ def trim_tts_output(
     return trimmed
 
 
+def preprocess_reference_audio(
+    audio: np.ndarray,
+    sample_rate: int,
+    peak_target: float = 0.95,
+    trim_top_db: float = 40.0,
+    edge_padding_ms: int = 100,
+) -> np.ndarray:
+    """
+    Clean up a reference-audio sample before validation/storage.
+
+    Removes DC offset, trims leading/trailing silence, and caps the peak so a
+    slightly-hot recording doesn't get rejected downstream as "clipping". The
+    goal is to accept reasonable real-world recordings — not to repair badly
+    distorted ones. True clipping artifacts inside the waveform can't be
+    recovered by peak scaling and will still sound bad.
+
+    Args:
+        audio: Mono audio array.
+        sample_rate: Sample rate of ``audio`` in Hz.
+        peak_target: Peak amplitude cap in [0, 1]. Applied only if the input
+            peak exceeds this value.
+        trim_top_db: Silence threshold for edge trimming, in dB below peak.
+            40 dB sits below normal speech dynamic range (≈30 dB) so soft
+            trailing syllables are preserved, while still catching obvious
+            leading/trailing silence. Lower values are more aggressive;
+            librosa's own default is 60.
+        edge_padding_ms: Milliseconds of padding to add back at each edge
+            *only if* trimming shortened the waveform, so TTS engines have a
+            brief silence to anchor on without ever making the output longer
+            than the input.
+
+    Returns:
+        Preprocessed audio array (float32).
+    """
+    audio = audio.astype(np.float32, copy=False)
+
+    if audio.size == 0:
+        return audio
+
+    audio = audio - float(np.mean(audio))
+
+    trimmed, _ = librosa.effects.trim(audio, top_db=trim_top_db)
+    if 0 < trimmed.size < audio.size:
+        pad_each = int(sample_rate * edge_padding_ms / 1000)
+        # Never pad past the original length — for near-max-duration uploads
+        # an unconditional pad would push them over the 30 s ceiling and
+        # trigger a spurious "too long" rejection.
+        headroom = (audio.size - trimmed.size) // 2
+        pad = min(pad_each, max(headroom, 0))
+        if pad > 0:
+            trimmed = np.pad(trimmed, (pad, pad), mode="constant")
+        audio = trimmed
+
+    peak = float(np.abs(audio).max())
+    if peak > peak_target and peak > 0:
+        audio = audio * (peak_target / peak)
+
+    return audio
+
+
 def validate_reference_audio(
     audio_path: str,
     min_duration: float = 2.0,
@@ -207,13 +267,13 @@ def validate_reference_audio(
 ) -> Tuple[bool, Optional[str]]:
     """
     Validate reference audio for voice cloning.
-    
+
     Args:
         audio_path: Path to audio file
         min_duration: Minimum duration in seconds
         max_duration: Maximum duration in seconds
         min_rms: Minimum RMS level
-        
+
     Returns:
         Tuple of (is_valid, error_message)
     """
@@ -231,26 +291,28 @@ def validate_and_load_reference_audio(
 ) -> Tuple[bool, Optional[str], Optional[np.ndarray], Optional[int]]:
     """
     Validate and load reference audio in a single pass.
-    
+
+    Applies :func:`preprocess_reference_audio` before checks so that
+    slightly-hot recordings aren't rejected as clipping. Duration and RMS
+    checks run on the preprocessed waveform.
+
     Returns:
         Tuple of (is_valid, error_message, audio_array, sample_rate)
     """
     try:
         audio, sr = load_audio(audio_path)
+        audio = preprocess_reference_audio(audio, sr)
         duration = len(audio) / sr
-        
+
         if duration < min_duration:
             return False, f"Audio too short (minimum {min_duration} seconds)", None, None
         if duration > max_duration:
             return False, f"Audio too long (maximum {max_duration} seconds)", None, None
-        
+
         rms = np.sqrt(np.mean(audio**2))
         if rms < min_rms:
             return False, "Audio is too quiet or silent", None, None
-        
-        if np.abs(audio).max() > 0.99:
-            return False, "Audio is clipping (reduce input gain)", None, None
-        
+
         return True, None, audio, sr
     except Exception as e:
         return False, f"Error validating audio: {str(e)}", None, None

backend/utils/hf_offline_patch.py

@@ -6,6 +6,7 @@ are already downloaded. Must be imported BEFORE mlx_audio.
 
 import logging
 import os
+import threading
 from contextlib import contextmanager
 from pathlib import Path
 from typing import Optional, Union
@@ -13,13 +14,33 @@ from typing import Optional, Union
 logger = logging.getLogger(__name__)
 
 
+# huggingface_hub reads ``HF_HUB_OFFLINE`` once at import time into
+# ``huggingface_hub.constants.HF_HUB_OFFLINE``; transformers mirrors that into
+# ``transformers.utils.hub._is_offline_mode`` at *its* import time. Toggling
+# ``os.environ`` after either module is imported does not flip those cached
+# bools, and the hot paths (``_http._default_backend_factory``,
+# ``transformers.utils.hub.is_offline_mode``) read the bools — not the env.
+# We mutate the cached constants directly, guarded by a refcount so
+# concurrent inference threads share a single offline window safely.
+
+_offline_lock = threading.RLock()
+_offline_refcount = 0
+_saved_env: Optional[str] = None
+_saved_hf_const: Optional[bool] = None
+_saved_transformers_const: Optional[bool] = None
+
+
 @contextmanager
 def force_offline_if_cached(is_cached: bool, model_label: str = ""):
-    """Context manager that sets ``HF_HUB_OFFLINE=1`` while loading a cached model.
+    """Force offline mode for the duration of a cached-model operation.
+
+    Flips ``HF_HUB_OFFLINE`` in the process env **and** in the cached bools
+    inside ``huggingface_hub.constants`` / ``transformers.utils.hub`` so HTTP
+    adapters and offline-mode checks actually see the change. Uses a refcount
+    so multiple concurrent inference threads share a single offline window
+    and the last one to exit restores state.
 
     If *is_cached* is ``False`` the block runs normally (network allowed).
-    If the offline load raises an error containing "offline" we automatically
-    retry with network access so a partially-cached model still works.
 
     Args:
         is_cached: Whether the model weights are already on disk.
@@ -29,34 +50,96 @@ def force_offline_if_cached(is_cached: bool, model_label: str = ""):
         yield
         return
 
-    original_value = os.environ.get("HF_HUB_OFFLINE")
-    os.environ["HF_HUB_OFFLINE"] = "1"
-    logger.info(
-        "[offline-guard] %s is cached — forcing HF_HUB_OFFLINE=1",
-        model_label or "model",
-    )
+    global _offline_refcount, _saved_env, _saved_hf_const, _saved_transformers_const
+
+    with _offline_lock:
+        if _offline_refcount == 0:
+            # Snapshot prior state, apply new state, roll back on *any*
+            # failure. Catching only ImportError here would let a partially
+            # broken install (RuntimeError, AttributeError from a half-init
+            # module, etc.) leave the cached HF constants mutated without
+            # bumping the refcount — a persistent offline leak that outlives
+            # the process and is miserable to debug.
+            prev_env = os.environ.get("HF_HUB_OFFLINE")
+            prev_hf: Optional[bool] = None
+            prev_tf: Optional[bool] = None
+            try:
+                try:
+                    import huggingface_hub.constants as hf_const
+
+                    prev_hf = hf_const.HF_HUB_OFFLINE
+                    hf_const.HF_HUB_OFFLINE = True
+                except ImportError:
+                    prev_hf = None
+
+                try:
+                    import transformers.utils.hub as tf_hub
+
+                    prev_tf = tf_hub._is_offline_mode
+                    tf_hub._is_offline_mode = True
+                except ImportError:
+                    prev_tf = None
+
+                os.environ["HF_HUB_OFFLINE"] = "1"
+            except BaseException:
+                # Roll back whatever we already changed, then re-raise so
+                # the caller sees the real failure.
+                if prev_hf is not None:
+                    try:
+                        import huggingface_hub.constants as hf_const
+
+                        hf_const.HF_HUB_OFFLINE = prev_hf
+                    except ImportError:
+                        pass
+                if prev_tf is not None:
+                    try:
+                        import transformers.utils.hub as tf_hub
+
+                        tf_hub._is_offline_mode = prev_tf
+                    except ImportError:
+                        pass
+                if prev_env is not None:
+                    os.environ["HF_HUB_OFFLINE"] = prev_env
+                else:
+                    os.environ.pop("HF_HUB_OFFLINE", None)
+                raise
+
+            _saved_env = prev_env
+            _saved_hf_const = prev_hf
+            _saved_transformers_const = prev_tf
+            logger.info(
+                "[offline-guard] %s is cached — forcing offline mode",
+                model_label or "model",
+            )
+        _offline_refcount += 1
 
     try:
         yield
-    except Exception as exc:
-        if "offline" in str(exc).lower():
-            logger.warning(
-                "[offline-guard] Offline load failed for %s, retrying with network: %s",
-                model_label or "model",
-                exc,
-            )
-            # Restore original env and retry — caller must wrap the load
-            # inside force_offline_if_cached so retrying here isn't possible.
-            # Instead, propagate a flag via the exception so the caller can
-            # decide.  For simplicity we just let it fall through to the
-            # finally block and re-raise.
-            raise
-        raise
     finally:
-        if original_value is not None:
-            os.environ["HF_HUB_OFFLINE"] = original_value
-        else:
-            os.environ.pop("HF_HUB_OFFLINE", None)
+        with _offline_lock:
+            _offline_refcount -= 1
+            if _offline_refcount == 0:
+                if _saved_env is not None:
+                    os.environ["HF_HUB_OFFLINE"] = _saved_env
+                else:
+                    os.environ.pop("HF_HUB_OFFLINE", None)
+                if _saved_hf_const is not None:
+                    try:
+                        import huggingface_hub.constants as hf_const
+
+                        hf_const.HF_HUB_OFFLINE = _saved_hf_const
+                    except ImportError:
+                        pass
+                if _saved_transformers_const is not None:
+                    try:
+                        import transformers.utils.hub as tf_hub
+
+                        tf_hub._is_offline_mode = _saved_transformers_const
+                    except ImportError:
+                        pass
+                _saved_env = None
+                _saved_hf_const = None
+                _saved_transformers_const = None
 
 
 def patch_huggingface_hub_offline():

bun.lock

@@ -17,7 +17,7 @@
     },
     "app": {
       "name": "@voicebox/app",
-      "version": "0.2.0",
+      "version": "0.4.1",
       "dependencies": {
         "@dnd-kit/core": "^6.3.1",
         "@dnd-kit/sortable": "^10.0.0",
@@ -72,9 +72,10 @@
     },
     "landing": {
       "name": "@voicebox/landing",
-      "version": "0.2.0",
+      "version": "0.4.1",
       "dependencies": {
         "@fontsource/space-grotesk": "^5.2.10",
+        "@icons-pack/react-simple-icons": "^13.13.0",
         "@radix-ui/react-separator": "^1.1.8",
         "@radix-ui/react-slot": "^1.2.4",
         "autoprefixer": "^10.4.17",
@@ -100,7 +101,7 @@
     },
     "tauri": {
       "name": "@voicebox/tauri",
-      "version": "0.2.0",
+      "version": "0.4.1",
       "dependencies": {
         "@tauri-apps/api": "^2.0.0",
         "@tauri-apps/plugin-dialog": "^2.0.0",
@@ -123,7 +124,7 @@
     },
     "web": {
       "name": "@voicebox/web",
-      "version": "0.2.0",
+      "version": "0.4.1",
       "dependencies": {
         "@tanstack/react-query": "^5.0.0",
         "react": "^18.3.0",
@@ -287,6 +288,8 @@
 
     "@humanwhocodes/object-schema": ["@humanwhocodes/[email protected]", "", {}, "sha512-93zYdMES/c1D69yZiKDBj0V24vqNzB/koF26KPaagAfd3P/4gUlh3Dys5ogAK+Exi9QyzlD8x/08Zt7wIKcDcA=="],
 
+    "@icons-pack/react-simple-icons": ["@icons-pack/[email protected]", "", { "peerDependencies": { "react": "^16.13 || ^17 || ^18 || ^19" } }, "sha512-B5HhQMIpcSH4z8IZ8HFhD59CboHceKYMpPC9kAwGyKntvPdyJJv26DLu4Z1wAjcCLyrJhf11tMhiQGom9Rxb9g=="],
+
     "@img/colour": ["@img/[email protected]", "", {}, "sha512-A5P/LfWGFSl6nsckYtjw9da+19jB8hkJ6ACTGcDfEJ0aE+l2n2El7dsVM7UVHZQ9s2lmYMWlrS21YLy2IR1LUw=="],
 
     "@img/sharp-darwin-arm64": ["@img/[email protected]", "", { "optionalDependencies": { "@img/sharp-libvips-darwin-arm64": "1.2.4" }, "os": "darwin", "cpu": "arm64" }, "sha512-imtQ3WMJXbMY4fxb/Ndp6HBTNVtWCUI0WdobyheGf5+ad6xX8VIDO8u2xE4qc/fr08CKG/7dDseFtn6M6g/r3w=="],

docs/PROJECT_STATUS.md

@@ -1,6 +1,6 @@
 # Voicebox Project Status & Roadmap
 
-> Last updated: 2026-04-18 | Current version: **v0.4.1** | ~155 open issues | 12 open PRs
+> Last updated: 2026-04-18 | Current version: **v0.4.1** | 232 open issues | 12 open PRs
 
 ---
 
@@ -224,6 +224,8 @@ POST /generate
 - **Blackwell (RTX 50-series) CUDA**: cu128 + sm_120 kernel support shipped (PR #401, #316), but users still report `cudaErrorNoKernelImageForDevice` (#417, #400, #396, #395, #390, #362) — likely a stale CUDA binary on upgraded installs. Needs a follow-up diagnostic / forced re-download path.
 - **Long text 50k character limit** (#464, #365, #354): Still hit on GPU despite chunking (PR #266). Chunking reliability needs another pass.
 - **ROCm on RDNA 3/4** (#469): `HSA_OVERRIDE_GFX_VERSION` is hardcoded and harms newer cards.
+- **`flash-attn is not installed` warning on every platform (cosmetic, common user complaint)**: Our transformer-based engines (Chatterbox / Qwen) emit `Warning: flash-attn is not installed. Will only run the manual PyTorch version. Please install flash-attn for faster inference.` on every startup, on every platform — we don't pin `flash-attn` in requirements because installing it is fragile and version-sensitive. Fallback is PyTorch SDPA, which is near-FA2 throughput on Ampere+ and is what actually runs. **Per-platform reality:** (a) **macOS/Apple Silicon** — FlashAttention is CUDA-only, irrelevant here; MLX has its own attention kernels. (b) **Linux** — `pip install flash-attn --no-build-isolation` works but takes 20+ min to compile. (c) **Windows** — no official support (Dao-AILab README still says only "Might work"; source builds routinely fail on recent CUDA/MSVC, issues #1715, #1828, #2395). Windows users can install community prebuilt wheels from `kingbri1/flash-attention` or `bdashore3/flash-attention` (latest v2.8.3, Aug 2025; `win_amd64` wheels for CUDA 12.4/12.8, Torch 2.6–2.9, Python 3.10–3.13) matching their exact CUDA/Torch/Python, or use WSL2. **Native-Windows alternatives worth considering as a build-time swap:** SageAttention (thu-ml, Apache 2.0, claims 2–5× over FA2) and xformers (official Windows wheels). **Action for us:** troubleshooting doc now covers it (see `docs/content/docs/overview/troubleshooting.mdx`), and we should optionally suppress the warning via `logging.getLogger(...).setLevel(ERROR)` at backend import since the fallback is functionally fine.
+- **WebAudio playback dies after audio-session interruption** (#41, plus an internal repro where the app is backgrounded long enough): WaveSurfer's `AudioContext` gets suspended by macOS — either because another app grabs the audio output, or because the WKWebView throttles when backgrounded. `play()` resolves and `timeupdate` can still fire, but no audio reaches the output. Only app restart fixes it. **Things already tried that didn't work:** (a) swapping WaveSurfer backend away from WebAudio — introduced more bugs, not an option; (b) remount hook on the player — doesn't help because a freshly-created `AudioContext` is born suspended and only resumes on a user gesture. PR #293 was a prior partial fix that doesn't cover this path. **Next thing to try** (not yet attempted — confirmed via grep of `AudioPlayer.tsx`): call `wavesurfer.getMediaElement().getGainNode().context.resume()` on the play button click (the click itself is a valid user gesture), plus a `visibilitychange` + `statechange` listener as belt-and-suspenders. The `ctx.resume()` pattern already exists in the codebase at `useStoryPlayback.ts:52` — just not wired into the main player.
 
 ---
 
@@ -303,9 +305,11 @@ POST /generate
 
 Still reported. Users get stuck downloads, can't resume, offline mode edge cases.
 
-**Key issues:** #475 (MAC CustomVoice install error), #449 (infinite loading macOS), #445 (can't download CustomVoice), #462 (Qwen requires internet even when loaded — regression from #150), #434 (infinite retry loop offline — PR #443 open), #432 (storage location change hangs when empty — partly fixed by PR #439/#433), #181, #180.
+**Key issues:** #475 (MAC CustomVoice install error), #449 (infinite loading macOS), #445 (can't download CustomVoice), #462 (Qwen requires internet even when loaded — regression from #150), #434 (infinite retry loop offline — PR #443 open), #432 (storage location change hangs when empty — partly fixed by PR #439/#433), #348 (TADA 3B Multilingual download fails), #336 (TADA model not listed in app), #275 (`No module named 'chatterbox'` on download), #304 (whisper-base feature extractor load error), #287 (macOS ARM `check_model_inputs` ImportError on new version), #181, #180.
 
-**Fix path:** PR #443 addresses infinite offline retry. CustomVoice-specific download failures (#475, #445) need triage — likely related to frozen-binary import fixes in PR #438.
+**Fix path:** PR #443 addresses infinite offline retry. CustomVoice-specific download failures (#475, #445) need triage — likely related to frozen-binary import fixes in PR #438. TADA cluster (#336, #348) and macOS ARM import regressions (#287, #275, #304) need a dedicated triage pass.
+
+**Qwen 0.6B-downloads-1.7B reports:** **#485** (2026-04-19), **#423** (macOS M1), **#329**. Originally a stale-fallback bug: `mlx-community/Qwen3-TTS-12Hz-0.6B-Base-bf16` wasn't published when MLX support shipped, so the 0.6B slot was aliased to the 1.7B repo. The 0.6B bf16 conversion is live now and both `backend/backends/mlx_backend.py` and `backend/backends/__init__.py` point at their correct repos. Qwen CustomVoice is unaffected — it runs via PyTorch on all platforms, both sizes always have dedicated repos.
 
 ### Language Requests (ongoing)
 
@@ -364,6 +368,9 @@ Notable:
 - **#383** — Concatenate partial reference audio into generated audio
 - **#382** — Lightning.ai support
 - **#376** — Remote mode
+- **#353** — Audio transcoding
+- **#317** — Voice pitch control
+- **#189** — "Auto" language option
 - **#173** — Vocal intonation/inflection control
 - **#165, #270** — Audiobook mode (PR #154 open)
 - **#242** — Seed value pinning
@@ -371,17 +378,40 @@ Notable:
 - **#235** — Finetuned Qwen3-TTS tokenizer (PR #253 open)
 - **#144** — Copy text to clipboard
 
+### Housekeeping / Triage Needed
+
+| Issue | Reason |
+|-------|--------|
+| **#431**, **#408** | Spam — Chinese "free Claude API" promos. Close. |
+| **#398** ("Excelente") | Non-issue. Close. |
+| **#357** | Informational — project featured in Awesome MLX. Close after acknowledgement. |
+| **#374**, **#377** | Version-release questions, no bug. Close. |
+| **#306** ("voice model"), **#389** ("New model"), **#473** ("New functionality") | Title-only issues, no content. Request details or close. |
+| **#309** | Uninstall/cleanup question. Answer and close. |
+| **#241** | "How to use in Colab" — support question, not a bug. |
+| **#423** / **#485** / **#329** | Stale MLX fallback to 1.7B repo — fixed; 0.6B bf16 conversion now live on `mlx-community`, registry points at correct repo on both backends. |
+| **#336** / **#348** | TADA download/registration cluster — triage together. |
+| **#287** / **#275** / **#304** | macOS ARM import regressions on new version — likely one root cause. |
+| **#292**, **#349** | Possibly already fixed by merged PRs (#321/#412 and #345). Verify + close. |
+
+**~70 older issues (pre-#170) not individually categorized above.** Most are long-tail support questions or duplicates of problems now addressed by the multi-engine / model-registry work. A dedicated backlog-sweep pass is overdue.
+
 ### Bugs (ongoing)
 
 | Category | Issues |
 |----------|--------|
-| Generation failures | #476, #467, #452, #459 (voice clone fetch error), #468 (tada-1b marked error), #437, #282 |
-| Audio quality | #456 (clipping errors v0.4.0), #436 (emotion labels), #333 (pitch/echo), #307 (by-model breakdown) |
-| File ops | #477 (spacy_pkuseg dict missing on frozen Windows build), #472 (storage location change) |
-| Windows | #466 (install problem), #273 (port 8000 conflict) |
-| Linux | #471 (thread-safe PULSE_SOURCE), #413 (Arch build), #409 (Kubuntu build), #341 |
-| macOS | #441 (older macOS), #369 (malware flag), #171 (ARM64 binary won't open) |
-| Profile/UI | #360 (Kokoro profile hides others — partly addressed by auto-switch), #299 (drag-drop on Win11), #329 (size selector state bug) |
+| Generation failures | #476, #467, #452, #459 (voice clone fetch error), #468 (tada-1b marked error), #437, #300, #301, #282 |
+| Audio quality | #456 (clipping errors v0.4.0), #436 (emotion labels), #333 (pitch/echo), #307 (by-model breakdown), #340 (all generations say "www...") |
+| Transcription | #371 (fails every time), #291 (extract transcription from generated audio) |
+| Effects / presets | #349 ("Failed to save" when creating effects presets — possibly fixed by merged #345) |
+| File ops | #477 (spacy_pkuseg dict missing on frozen Windows build), #472 (storage location change), #283 (allow longer files for voice creation + in-app trim), #350 (failed to add sample) |
+| History | #292 (can't delete failed generations — possibly fixed by merged #321/#412) |
+| Windows | #466 (install problem), #375 (WinError 5 access denied), #273 (port 8000 conflict), #201 (model doesn't stay loaded) |
+| Linux | #471 (thread-safe PULSE_SOURCE), #413 (Arch build), #409 (Kubuntu build), #351, #341 |
+| macOS | #441 (older macOS), #369 (malware flag), #334 (microphone permission), #287 (`check_model_inputs` ImportError — regression), #171 (ARM64 binary won't open) |
+| Profile/UI | #360 (Kokoro profile hides others — partly addressed by auto-switch), #299 (drag-drop on Win11), #329 (size selector state bug), #393 (stuck loading screen after reinstall to new dir) |
+| Integrations | #397 (SAMMI-bot 422 Unprocessable Entity) |
+| Audio playback / session | **#41** (macOS: Voicebox goes silent after another app takes audio output; restart restores it) — see deep-dive below |
 | Database | #174 (sqlite3 IntegrityError) |
 
 ---

docs/content/docs/TROUBLESHOOTING.md

@@ -1,311 +0,0 @@
----
-title: "Troubleshooting Guide"
-description: "Common issues and solutions for Voicebox"
----
-
-Common issues and solutions for Voicebox.
-
-## Installation Issues
-
-### macOS: "Voicebox cannot be opened because it is from an unidentified developer"
-
-**Solution:**
-1. Right-click the `.dmg` file
-2. Select "Open"
-3. Click "Open" in the security dialog
-4. Alternatively, go to System Settings → Privacy & Security → Allow Voicebox
-
-### Windows: "Windows protected your PC"
-
-**Solution:**
-1. Click "More info"
-2. Click "Run anyway"
-3. Windows Defender may flag new software; this is normal for unsigned apps
-
-### Linux: AppImage won't run
-
-**Solution:**
-```bash
-chmod +x voicebox-*.AppImage
-./voicebox-*.AppImage
-```
-
-## Runtime Issues
-
-### Server won't start
-
-**Symptoms:** App opens but shows "Server not connected"
-
-**Solutions:**
-1. **Check Python installation**
-   ```bash
-   python --version  # Should be 3.11+
-   ```
-
-2. **Check server binary exists**
-   - Look in `tauri/src-tauri/binaries/` for your platform
-   - Binary should match your system architecture
-
-3. **Check permissions**
-   ```bash
-   # macOS/Linux
-   chmod +x tauri/src-tauri/binaries/voicebox-server-*
-   ```
-
-4. **Check logs**
-   - macOS: Open Console.app and search for "voicebox"
-   - Linux: Check `~/.local/share/voicebox/` for logs
-   - Windows: Check Event Viewer
-
-### "Model download failed"
-
-**Symptoms:** First generation fails with download error
-
-**Solutions:**
-1. **Check internet connection**
-   - Models download from HuggingFace Hub (~2-4GB)
-   - First download may take several minutes
-
-2. **Check disk space**
-   - Models are cached in `~/.cache/huggingface/`
-   - Ensure at least 5GB free space
-
-3. **Manual download** (if automatic fails)
-   ```bash
-   pip install huggingface_hub
-   huggingface-cli download Qwen/Qwen3-TTS-12Hz-1.7B-Base
-   ```
-
-### "Out of memory" errors
-
-**Symptoms:** Generation fails with CUDA/VRAM errors
-
-**Solutions:**
-1. **Use smaller model**
-   - Switch to 0.6B model instead of 1.7B
-   - Settings → Model Management → Load 0.6B
-
-2. **Close other applications**
-   - Free up GPU memory
-   - Close browser tabs, other ML apps
-
-3. **Use CPU mode**
-   - Slower but works without GPU
-   - Backend automatically falls back to CPU
-
-### MLX "Failed to load the default metallib" error (Apple Silicon)
-
-**Symptoms:** Generation fails with "library not found" or "metallib" errors
-
-**Solutions:**
-1. **Rebuild server binary**
-   ```bash
-   just build-server
-   ```
-   The build script automatically includes MLX Metal shader libraries on Apple Silicon.
-
-2. **Check MLX installation**
-   ```bash
-   pip install -r backend/requirements-mlx.txt
-   ```
-
-3. **Verify backend detection**
-   - Check server logs for "Backend: MLX"
-   - If showing "Backend: PYTORCH", MLX may not be installed correctly
-
-### Audio playback issues
-
-**Symptoms:** Generated audio won't play
-
-**Solutions:**
-1. **Check audio format**
-   - Audio is saved as WAV files
-   - Ensure your system supports WAV playback
-
-2. **Try downloading audio**
-   - Right-click → Download
-   - Play in external player
-
-3. **Check browser permissions** (web version)
-   - Allow audio autoplay in browser settings
-
-### Slow generation
-
-**Symptoms:** Generation takes >30 seconds
-
-**Solutions:**
-1. **Check backend type** (Apple Silicon)
-   - Check Settings → Server Status
-   - Should show "Backend: MLX" on Apple Silicon
-   - If showing "Backend: PYTORCH", install MLX: `pip install -r backend/requirements-mlx.txt`
-   - MLX provides 4-5x faster inference on Apple Silicon
-
-2. **Use GPU** (if available)
-   - Check Settings → Server Status
-   - Should show "GPU available: true"
-   - Apple Silicon: Should show "Metal (Apple Silicon via MLX)"
-   - Windows/Linux: Should show "CUDA" if GPU available
-
-3. **Enable caching**
-   - Voice prompts are cached automatically
-   - Second generation with same voice should be faster
-
-4. **Use smaller model**
-   - 0.6B model is faster than 1.7B
-   - Quality difference is minimal for most voices
-
-5. **Check system resources**
-   - Close other CPU/GPU intensive apps
-   - Ensure adequate RAM (8GB+ recommended)
-
-## API Issues
-
-### "Connection refused" when using API
-
-**Solutions:**
-1. **Check server is running**
-   ```bash
-   curl http://localhost:17493/health
-   ```
-
-2. **Check remote mode**
-   - If connecting remotely, ensure server is started with `--host 0.0.0.0`
-   - Check firewall settings
-
-3. **Check port availability**
-   - The current local app and dev workflow uses port 17493 by default
-   - Ensure no other service is using it
-
-### CORS errors in browser
-
-**Solutions:**
-1. **Use desktop app** (recommended)
-   - Desktop app doesn't have CORS restrictions
-
-2. **Configure CORS** (for web deployment)
-   - Update `backend/main.py` CORS settings
-   - Add your domain to allowed origins
-
-## Update Issues
-
-### "Update check failed"
-
-**Solutions:**
-1. **Check internet connection**
-   - Updates are fetched from GitHub releases
-
-2. **Check GitHub access**
-   - Ensure `github.com` is accessible
-   - Check firewall/proxy settings
-
-3. **Manual update**
-   - Download latest release from GitHub
-   - Install manually
-
-### "Invalid signature" error
-
-**Solutions:**
-1. **Re-download installer**
-   - Signature may be corrupted
-   - Download fresh copy from GitHub
-
-2. **Check release integrity**
-   - Verify `.sig` file matches installer
-   - Report issue if signature is invalid
-
-## Data Issues
-
-### Profiles disappeared
-
-**Solutions:**
-1. **Check data directory**
-   - macOS: `~/Library/Application Support/sh.voicebox.app/`
-   - Windows: `%APPDATA%/sh.voicebox.app/`
-   - Linux: `~/.config/sh.voicebox.app/`
-
-2. **Check database**
-   - Database: `data/voicebox.db`
-   - Ensure file exists and is readable
-
-3. **Restore from backup**
-   - Profiles can be exported/imported
-   - Check for backup files
-
-### "Database locked" error
-
-**Solutions:**
-1. **Close other instances**
-   - Ensure only one Voicebox instance is running
-
-2. **Restart app**
-   - Close and reopen Voicebox
-
-3. **Check file permissions**
-   - Ensure database file is writable
-   - Check directory permissions
-
-## Development Issues
-
-### Build fails
-
-**Solutions:**
-1. **Check Rust installation**
-   ```bash
-   rustc --version
-   rustup update
-   ```
-
-2. **Check Tauri dependencies**
-   ```bash
-   cd tauri
-   bun install
-   ```
-
-3. **Clean build**
-   ```bash
-   cd tauri/src-tauri
-   cargo clean
-   cd ../..
-   just build
-   ```
-
-### API client generation fails
-
-**Solutions:**
-1. **Start backend server**
-   ```bash
-   just dev-backend
-   ```
-
-2. **Check OpenAPI endpoint**
-   ```bash
-   curl http://localhost:17493/openapi.json
-   ```
-
-3. **Regenerate client**
-   ```bash
-   just generate-api
-   ```
-
-## Still Having Issues?
-
-1. **Check existing issues**
-   - Search GitHub issues for similar problems
-   - Check closed issues for solutions
-
-2. **Create new issue**
-   - Include:
-     - OS and version
-     - Voicebox version
-     - Steps to reproduce
-     - Error messages/logs
-     - Screenshots (if applicable)
-
-3. **Get help**
-   - Check documentation in `docs/`
-   - Review `backend/README.md` for API details
-   - See `CONTRIBUTING.md` for development help
-
----
-
-For more help, open an issue on [GitHub](https://github.com/jamiepine/voicebox/issues).

docs/content/docs/overview/troubleshooting.mdx

@@ -29,6 +29,14 @@ Windows SmartScreen may warn that the app is unrecognized.
   This is expected for unsigned applications. We're working on code signing for future releases.
 </Callout>
 
+### Linux: AppImage Won't Run
+
+**Solution:**
+```bash
+chmod +x voicebox-*.AppImage
+./voicebox-*.AppImage
+```
+
 ## Server Issues
 
 ### Backend Server Won't Start
@@ -85,6 +93,52 @@ Windows SmartScreen may warn that the app is unrecognized.
   </Accordion>
 </AccordionGroup>
 
+### `flash-attn is not installed` Warning in Server Logs
+
+**Symptoms:**
+```
+Warning: flash-attn is not installed. Will only run the manual PyTorch version.
+Please install flash-attn for faster inference.
+```
+
+**This is harmless.** The warning is emitted by our transformer-based engines (Chatterbox / Qwen) on every startup. FlashAttention is an optional acceleration library — when it's not present, PyTorch's built-in scaled-dot-product attention (SDPA) runs instead, which is near-FA2 throughput on modern GPUs. Generation works normally.
+
+**Why it shows up on every platform:**
+- **Windows:** `flash-attn` has no official Windows support. The upstream project (Dao-AILab/flash-attention) still only says it *might* work, and source builds typically fail on recent CUDA/MSVC combinations.
+- **macOS (Apple Silicon):** FlashAttention is CUDA-only and doesn't apply here at all. MLX has its own optimized attention kernels.
+- **Linux:** It's not pinned in our requirements because installing it is fragile and version-sensitive; users who want it install it themselves.
+
+**Solutions (all optional):**
+
+<AccordionGroup>
+  <Accordion title="Ignore it (recommended)">
+    PyTorch SDPA is what actually runs the model, and on Ampere/Ada/Hopper GPUs it's within a few percent of FA2 for our workloads. You won't notice a meaningful speed difference.
+  </Accordion>
+
+  <Accordion title="Install flash-attn on Linux">
+    ```bash
+    pip install flash-attn --no-build-isolation
+    ```
+
+    Requires a matching CUDA toolkit. Build can take 20+ minutes.
+  </Accordion>
+
+  <Accordion title="Install flash-attn on Windows (community wheels)">
+    Official builds don't exist, but community maintainers publish prebuilt wheels:
+
+    - [kingbri1/flash-attention releases](https://github.com/kingbri1/flash-attention/releases)
+    - [bdashore3/flash-attention releases](https://github.com/bdashore3/flash-attention/releases)
+
+    Pick the wheel matching your exact CUDA + PyTorch + Python combination. Example:
+
+    ```bash
+    pip install https://github.com/kingbri1/flash-attention/releases/download/v2.8.3/flash_attn-2.8.3+cu128torch2.8.0cxx11abiFALSE-cp312-cp312-win_amd64.whl
+    ```
+
+    Alternatively, run Voicebox's backend inside WSL2 and use the standard Linux wheels.
+  </Accordion>
+</AccordionGroup>
+
 ### Connection Timeout
 
 **Symptoms:**
@@ -174,6 +228,34 @@ This is expected behavior. The first generation downloads the selected TTS engin
   </Accordion>
 </AccordionGroup>
 
+### MLX "Failed to load the default metallib" (Apple Silicon)
+
+**Symptoms:**
+- Generation fails with "library not found" or "metallib" errors
+- Server logs reference missing Metal shader libraries
+
+**Solutions:**
+
+<AccordionGroup>
+  <Accordion title="Rebuild the Server Binary">
+    ```bash
+    just build-server
+    ```
+
+    The build script bundles MLX Metal shader libraries on Apple Silicon automatically.
+  </Accordion>
+
+  <Accordion title="Reinstall MLX Dependencies">
+    ```bash
+    pip install -r backend/requirements-mlx.txt
+    ```
+  </Accordion>
+
+  <Accordion title="Verify Backend Detection">
+    Check Settings → Server Status. Should show **Backend: MLX** on Apple Silicon. If it shows **Backend: PYTORCH**, MLX isn't installed correctly.
+  </Accordion>
+</AccordionGroup>
+
 ## Audio Issues
 
 ### No Audio Playback
@@ -357,7 +439,12 @@ Restart the app to create a fresh database.
 - Check your internet connection
 - Check HuggingFace Hub status
 - Try using a VPN if HuggingFace is blocked in your region
-- Manually download and place in cache directory
+- Manually download via the HuggingFace CLI and place in the cache directory:
+
+  ```bash
+  pip install huggingface_hub
+  huggingface-cli download Qwen/Qwen3-TTS-12Hz-1.7B-Base
+  ```
 
 ### Wrong Model Version
 
@@ -404,6 +491,16 @@ rmdir /s %USERPROFILE%\.cache\huggingface\hub\models--Qwen*
   <Accordion title="Update GPU Drivers">
     Outdated drivers can cause performance issues. Update to the latest NVIDIA drivers.
   </Accordion>
+
+  <Accordion title="Apple Silicon: Confirm MLX Backend">
+    Check Settings → Server Status. Should show **Backend: MLX** on Apple Silicon — MLX is 4–5× faster than PyTorch here. If it shows **Backend: PYTORCH**, reinstall MLX:
+
+    ```bash
+    pip install -r backend/requirements-mlx.txt
+    ```
+
+    GPU availability should read "Metal (Apple Silicon via MLX)".
+  </Accordion>
 </AccordionGroup>
 
 ### High Memory Usage
@@ -417,6 +514,21 @@ rmdir /s %USERPROFILE%\.cache\huggingface\hub\models--Qwen*
 - Clear generation history
 - Restart the app periodically
 
+## Update Issues
+
+### "Update Check Failed"
+
+**Solutions:**
+- Confirm your internet connection — updates are fetched from GitHub releases.
+- Ensure `github.com` is accessible and not blocked by a firewall or proxy.
+- As a fallback, download the latest release from GitHub and install manually.
+
+### "Invalid Signature" Error
+
+**Solutions:**
+- Re-download the installer — the signature may have been corrupted in transit.
+- Verify the `.sig` file matches the installer; if it doesn't, file an issue.
+
 ## Remote Mode Issues
 
 ### Can't Connect to Remote Server
@@ -482,5 +594,3 @@ python --version
 # GPU info (if generation issues)
 nvidia-smi  # NVIDIA GPUs
 ```
-
-For more detailed troubleshooting, see the [TROUBLESHOOTING.md](https://github.com/jamiepine/voicebox/blob/main/docs/TROUBLESHOOTING.md) file in the repository.

landing/package.json

@@ -1,15 +1,16 @@
 {
   "name": "@voicebox/landing",
-  "version": "0.4.1",
+  "version": "0.4.2",
   "description": "Landing page for voicebox.sh",
   "scripts": {
-    "dev": "bun --bun next dev --turbo",
+    "dev": "next dev --turbo",
     "build": "bun --bun next build",
     "start": "bun --bun next start",
     "lint": "next lint"
   },
   "dependencies": {
     "@fontsource/space-grotesk": "^5.2.10",
+    "@icons-pack/react-simple-icons": "^13.13.0",
     "@radix-ui/react-separator": "^1.1.8",
     "@radix-ui/react-slot": "^1.2.4",
     "autoprefixer": "^10.4.17",

landing/src/app/download/[platform]/route.ts

@@ -1,42 +1,46 @@
 import { type NextRequest, NextResponse } from 'next/server';
-import { getLatestRelease } from '@/lib/releases';
 
 export const dynamic = 'force-dynamic';
 
-const PLATFORM_MAP: Record<
-  string,
-  keyof Awaited<ReturnType<typeof getLatestRelease>>['downloadLinks']
-> = {
+// Pretty URLs from README / docs (e.g. /download/mac-arm) are kept for
+// compatibility, but we now always route through the /download page so users
+// see context + a donate prompt + resources while the download kicks off.
+// The page handles the actual file trigger itself — no more silent redirects
+// to GitHub or direct asset URLs.
+const PLATFORM_ALIAS: Record<string, string> = {
   'mac-arm': 'macArm',
+  macArm: 'macArm',
   'mac-intel': 'macIntel',
+  macIntel: 'macIntel',
   windows: 'windows',
-  linux: 'linux',
 };
 
+function getPublicOrigin(request: NextRequest): string {
+  const forwardedHost = request.headers.get('x-forwarded-host');
+  const forwardedProto = request.headers.get('x-forwarded-proto');
+
+  if (forwardedHost && forwardedProto) {
+    // Behind reverse proxies/CDNs, request.url can be an internal origin
+    // (for example localhost:8080). Prefer forwarded headers so redirects
+    // keep users on the public domain.
+    return `${forwardedProto}://${forwardedHost}`;
+  }
+
+  return new URL(request.url).origin;
+}
+
 export async function GET(
-  _request: NextRequest,
+  request: NextRequest,
   { params }: { params: Promise<{ platform: string }> },
 ) {
+  const origin = getPublicOrigin(request);
   const { platform } = await params;
-  const key = PLATFORM_MAP[platform];
-
-  if (!key) {
-    return NextResponse.json(
-      { error: `Unknown platform: ${platform}. Use: ${Object.keys(PLATFORM_MAP).join(', ')}` },
-      { status: 404 },
-    );
-  }
-
-  try {
-    const release = await getLatestRelease();
-    const url = release.downloadLinks[key];
-
-    if (!url) {
-      return NextResponse.json({ error: `No download available for ${platform}` }, { status: 404 });
-    }
-
-    return NextResponse.redirect(url);
-  } catch {
-    return NextResponse.redirect(`https://github.com/jamiepine/voicebox/releases/latest`);
+  // No prebuilt Linux binary yet — send straight to the build-from-source page.
+  if (platform === 'linux') {
+    return NextResponse.redirect(new URL('/linux-install', origin), 307);
   }
+  const normalized = PLATFORM_ALIAS[platform];
+  const target = new URL('/download', origin);
+  if (normalized) target.searchParams.set('platform', normalized);
+  return NextResponse.redirect(target, 307);
 }

landing/src/app/download/page.tsx

@@ -0,0 +1,313 @@
+'use client';
+
+import {
+  ArrowLeft,
+  Bot,
+  Coffee,
+  Download as DownloadIcon,
+  FileText,
+  Github,
+} from 'lucide-react';
+import Image from 'next/image';
+import Link from 'next/link';
+import { useEffect, useMemo, useState } from 'react';
+import { AppleIcon, LinuxIcon, WindowsIcon } from '@/components/PlatformIcons';
+import { Button } from '@/components/ui/button';
+import { DONATE_URL, GITHUB_RELEASES_PAGE, GITHUB_REPO } from '@/lib/constants';
+import type { DownloadLinks } from '@/lib/releases';
+
+type Platform = keyof DownloadLinks;
+
+type PlatformMeta = {
+  key: Platform;
+  label: string;
+  description: string;
+  icon: React.ComponentType<{ className?: string }>;
+};
+
+const PLATFORMS: PlatformMeta[] = [
+  { key: 'macArm', label: 'macOS', description: 'Apple Silicon', icon: AppleIcon },
+  { key: 'macIntel', label: 'macOS', description: 'Intel (x64)', icon: AppleIcon },
+  { key: 'windows', label: 'Windows', description: '64-bit (MSI)', icon: WindowsIcon },
+  { key: 'linux', label: 'Linux', description: 'Build from source', icon: LinuxIcon },
+];
+
+function detectPlatform(): Platform | null {
+  if (typeof navigator === 'undefined') return null;
+  const ua = navigator.userAgent;
+  if (/Windows/i.test(ua)) return 'windows';
+  if (/Linux/i.test(ua) && !/Android/i.test(ua)) return 'linux';
+  if (/Mac/i.test(ua)) {
+    // Apple Silicon Safari reports "Intel" for compat; default to ARM since
+    // M-series is the majority. Users can click the Intel button if needed.
+    return 'macArm';
+  }
+  return null;
+}
+
+function parseQueryPlatform(search: string): Platform | null {
+  const params = new URLSearchParams(search);
+  const raw = params.get('platform');
+  if (!raw) return null;
+  // Accept both camelCase and hyphenated forms (/download/mac-arm → ?platform=mac-arm).
+  const normalized = raw
+    .toLowerCase()
+    .replace(/[-_\s]/g, '')
+    .replace('macarm', 'macArm')
+    .replace('macintel', 'macIntel');
+  const valid: Platform[] = ['macArm', 'macIntel', 'windows', 'linux'];
+  return (valid as string[]).includes(normalized) ? (normalized as Platform) : null;
+}
+
+export default function DownloadPage() {
+  const [links, setLinks] = useState<DownloadLinks | null>(null);
+  const [linksError, setLinksError] = useState(false);
+  const [platform, setPlatform] = useState<Platform | null>(null);
+  const [triggered, setTriggered] = useState(false);
+
+  useEffect(() => {
+    const fromQuery = parseQueryPlatform(window.location.search);
+    const resolved = fromQuery ?? detectPlatform();
+    // No prebuilt Linux binary yet — send Linux users to the build-from-source
+    // instructions instead of sitting on /download trying to trigger a
+    // download that doesn't exist.
+    if (resolved === 'linux') {
+      window.location.replace('/linux-install');
+      return;
+    }
+    setPlatform(resolved);
+  }, []);
+
+  useEffect(() => {
+    let cancelled = false;
+    fetch('/api/releases')
+      .then((r) => {
+        if (!r.ok) throw new Error(`releases ${r.status}`);
+        return r.json();
+      })
+      .then((data) => {
+        if (cancelled) return;
+        if (data.downloadLinks) setLinks(data.downloadLinks as DownloadLinks);
+      })
+      .catch(() => {
+        if (!cancelled) setLinksError(true);
+      });
+    return () => {
+      cancelled = true;
+    };
+  }, []);
+
+  useEffect(() => {
+    if (triggered || !links || !platform) return;
+    const url = links[platform];
+    if (!url) return;
+
+    const a = document.createElement('a');
+    a.href = url;
+    a.rel = 'noopener';
+    a.style.display = 'none';
+    document.body.appendChild(a);
+    a.click();
+    document.body.removeChild(a);
+    setTriggered(true);
+  }, [triggered, links, platform]);
+
+  const activeMeta = useMemo(
+    () => PLATFORMS.find((p) => p.key === platform) ?? null,
+    [platform],
+  );
+
+  return (
+    <div className="min-h-screen bg-background">
+      {/* Minimal branded header */}
+      <header className="border-b border-border/50">
+        <div className="mx-auto flex max-w-5xl items-center justify-between px-6 py-4">
+          <Link href="/" className="flex items-center gap-2.5">
+            <Image
+              src="/voicebox-logo-app.webp"
+              alt="Voicebox"
+              width={28}
+              height={28}
+              className="h-7 w-7"
+            />
+            <span className="text-[15px] font-semibold text-foreground">Voicebox</span>
+          </Link>
+          <Link
+            href="/"
+            className="flex items-center gap-1.5 text-sm text-muted-foreground hover:text-foreground transition-colors"
+          >
+            <ArrowLeft className="h-3.5 w-3.5" />
+            Back to voicebox.sh
+          </Link>
+        </div>
+      </header>
+
+      <main className="mx-auto max-w-5xl px-6 py-16 md:py-24">
+        {/* Hero */}
+        <div className="flex flex-col md:flex-row md:items-center gap-10 md:gap-14">
+          <Image
+            src="/voicebox-logo-app.webp"
+            alt="Voicebox"
+            width={200}
+            height={200}
+            priority
+            className="h-32 w-32 md:h-44 md:w-44 shrink-0 drop-shadow-2xl"
+          />
+          <div className="flex-1 min-w-0 text-center md:text-left">
+            {triggered ? (
+              <>
+                <h1 className="text-4xl md:text-5xl font-semibold tracking-tight text-foreground mb-4">
+                  Your download has started.
+                </h1>
+                <p className="text-lg text-muted-foreground">
+                  {activeMeta
+                    ? `Downloading Voicebox for ${activeMeta.label} (${activeMeta.description}). Check your downloads folder.`
+                    : 'Check your downloads folder for Voicebox.'}
+                </p>
+              </>
+            ) : (
+              <>
+                <h1 className="text-4xl md:text-5xl font-semibold tracking-tight text-foreground mb-4">
+                  {linksError ? "We couldn't load the latest release." : 'Download Voicebox'}
+                </h1>
+                <p className="text-lg text-muted-foreground">
+                  {linksError
+                    ? 'Our release server is temporarily unreachable. Please try again in a moment.'
+                    : 'Pick your platform to get started.'}
+                </p>
+              </>
+            )}
+          </div>
+        </div>
+
+        {/* Platform buttons — always visible as a fallback */}
+        {linksError ? (
+          <div className="mt-12 rounded-xl border border-border bg-card/60 backdrop-blur-sm p-6 text-center">
+            <p className="text-sm text-muted-foreground mb-4">
+              If this keeps happening, you can{' '}
+              <a
+                href={`${GITHUB_RELEASES_PAGE}/latest`}
+                target="_blank"
+                rel="noopener noreferrer"
+                className="text-accent underline underline-offset-2 hover:text-accent/80"
+              >
+                browse releases on GitHub
+              </a>
+              {' '}and grab the build for your platform manually.
+            </p>
+          </div>
+        ) : (
+          <div className="mt-12 rounded-xl border border-border bg-card/60 backdrop-blur-sm p-6">
+            <h2 className="text-sm font-medium text-foreground mb-4">
+              {triggered ? 'Download not working?' : 'Choose your platform'}
+            </h2>
+            <div className="grid grid-cols-1 sm:grid-cols-2 gap-3">
+              {PLATFORMS.map((meta) => {
+                const isLinux = meta.key === 'linux';
+                const url = isLinux ? '/linux-install' : links?.[meta.key];
+                const isActive = meta.key === platform;
+                const disabled = !isLinux && !url;
+                return (
+                  <a
+                    key={meta.key}
+                    href={url ?? '#'}
+                    {...(isLinux ? {} : { download: true })}
+                    aria-disabled={disabled}
+                    onClick={(e) => {
+                      if (disabled) e.preventDefault();
+                    }}
+                    className={`flex items-center rounded-xl border px-5 py-4 transition-all group ${
+                      isActive
+                        ? 'border-accent/40 bg-accent/5 hover:border-accent/60'
+                        : 'border-border bg-card/40 hover:border-accent/30 hover:bg-card'
+                    } ${disabled ? 'opacity-50 cursor-not-allowed' : ''}`}
+                  >
+                    <meta.icon className="h-6 w-6 shrink-0 text-muted-foreground group-hover:text-foreground transition-colors" />
+                    <div className="ml-4 flex-1">
+                      <div className="text-sm font-medium text-foreground">{meta.label}</div>
+                      <div className="text-xs text-muted-foreground">{meta.description}</div>
+                    </div>
+                    <DownloadIcon className="h-4 w-4 text-muted-foreground/60 group-hover:text-accent transition-colors" />
+                  </a>
+                );
+              })}
+            </div>
+          </div>
+        )}
+
+        {/* Donate — prominent, heartfelt, post-click context */}
+        <div className="mt-16 rounded-2xl border border-border bg-gradient-to-br from-card via-card/80 to-background backdrop-blur-sm p-8 md:p-10 overflow-hidden relative">
+          <div className="absolute top-0 right-0 w-64 h-64 bg-[#FFDD00]/5 rounded-full blur-3xl -translate-y-1/2 translate-x-1/2 pointer-events-none" />
+          <div className="relative">
+            <div className="inline-flex items-center gap-2 rounded-full border border-[#FFDD00]/30 bg-[#FFDD00]/10 px-3 py-1 mb-4">
+              <Coffee className="h-3 w-3 text-[#FFDD00]" />
+              <span className="text-[11px] font-medium uppercase tracking-wider text-[#FFDD00]">
+                Hi from the maintainer
+              </span>
+            </div>
+            <h2 className="text-2xl md:text-3xl font-semibold tracking-tight text-foreground mb-4">
+              Jamie here — Voicebox is a side project.
+            </h2>
+            <p className="text-muted-foreground leading-relaxed mb-6 max-w-2xl">
+              I build and maintain Voicebox in my spare time. It's completely
+              free, open source, runs entirely on your machine — no accounts, no
+              cloud, no subscriptions, no upsells. If it saves you an ElevenLabs
+              bill or just made your day, a coffee genuinely helps me keep
+              shipping updates, adding new models, and fixing bugs. Every little
+              bit keeps the lights on.
+            </p>
+            <Button asChild size="lg" className="bg-[#FFDD00]/10 border-[#FFDD00]/30 text-[#FFDD00] hover:bg-[#FFDD00]/20 hover:border-[#FFDD00]/50">
+              <a href={DONATE_URL} target="_blank" rel="noopener noreferrer">
+                <Coffee className="h-4 w-4 mr-2" />
+                Buy me a coffee
+              </a>
+            </Button>
+          </div>
+        </div>
+
+        {/* Resources */}
+        <div className="mt-10">
+          <h2 className="text-sm font-medium text-foreground mb-4">While you wait</h2>
+          <div className="grid grid-cols-1 md:grid-cols-3 gap-4">
+            <a
+              href="https://docs.voicebox.sh"
+              target="_blank"
+              rel="noopener noreferrer"
+              className="rounded-xl border border-border bg-card/60 backdrop-blur-sm p-5 hover:border-accent/30 hover:bg-card transition-all group"
+            >
+              <FileText className="h-5 w-5 text-accent mb-3" />
+              <h3 className="text-sm font-medium text-foreground mb-1">Read the docs</h3>
+              <p className="text-xs text-muted-foreground leading-relaxed">
+                Get familiar with Voicebox — setup, voice cloning, the REST API.
+              </p>
+            </a>
+            <a
+              href="https://deepwiki.com/jamiepine/voicebox"
+              target="_blank"
+              rel="noopener noreferrer"
+              className="rounded-xl border border-border bg-card/60 backdrop-blur-sm p-5 hover:border-accent/30 hover:bg-card transition-all group"
+            >
+              <Bot className="h-5 w-5 text-accent mb-3" />
+              <h3 className="text-sm font-medium text-foreground mb-1">Got questions? Ask AI.</h3>
+              <p className="text-xs text-muted-foreground leading-relaxed">
+                DeepWiki is an AI that knows Voicebox inside-out. Ask anything.
+              </p>
+            </a>
+            <a
+              href={GITHUB_REPO}
+              target="_blank"
+              rel="noopener noreferrer"
+              className="rounded-xl border border-border bg-card/60 backdrop-blur-sm p-5 hover:border-accent/30 hover:bg-card transition-all group"
+            >
+              <Github className="h-5 w-5 text-accent mb-3" />
+              <h3 className="text-sm font-medium text-foreground mb-1">Source on GitHub</h3>
+              <p className="text-xs text-muted-foreground leading-relaxed">
+                Star the repo, file issues, or contribute a PR.
+              </p>
+            </a>
+          </div>
+        </div>
+      </main>
+    </div>
+  );
+}

landing/src/app/page.tsx

@@ -17,12 +17,9 @@ import {Navbar} from "@/components/Navbar";
 import {AppleIcon, LinuxIcon, WindowsIcon} from "@/components/PlatformIcons";
 import {TutorialsSection} from "@/components/TutorialsSection";
 import {VoiceCreator} from "@/components/VoiceCreator";
-import {DOWNLOAD_LINKS, GITHUB_REPO} from "@/lib/constants";
-import type {DownloadLinks} from "@/lib/releases";
+import {GITHUB_REPO} from "@/lib/constants";
 
 export default function Home() {
-	const [downloadLinks, setDownloadLinks] =
-		useState<DownloadLinks>(DOWNLOAD_LINKS);
 	const [version, setVersion] = useState<string | null>(null);
 	const [totalDownloads, setTotalDownloads] = useState<number | null>(null);
 
@@ -33,7 +30,6 @@ export default function Home() {
 				return res.json();
 			})
 			.then((data) => {
-				if (data.downloadLinks) setDownloadLinks(data.downloadLinks);
 				if (data.version) setVersion(data.version);
 				if (data.totalDownloads != null) setTotalDownloads(data.totalDownloads);
 			})
@@ -92,7 +88,7 @@ export default function Home() {
 						style={{animationDelay: "300ms"}}
 					>
 						<a
-							href="#download"
+							href="/download"
 							className="rounded-full bg-accent px-8 py-3.5 text-sm font-semibold uppercase tracking-wider text-white shadow-[0_4px_20px_hsl(43_60%_50%/0.3),inset_0_2px_0_rgba(255,255,255,0.2),inset_0_-2px_0_rgba(0,0,0,0.1)] transition-all hover:bg-accent-faint active:shadow-[0_2px_10px_hsl(43_60%_50%/0.3),inset_0_4px_8px_rgba(0,0,0,0.3)]"
 						>
 							Download
@@ -403,8 +399,7 @@ export default function Home() {
 					<div className="grid grid-cols-1 sm:grid-cols-2 gap-3 max-w-2xl mx-auto">
 						{/* macOS ARM */}
 						<a
-							href={downloadLinks.macArm}
-							download
+							href="/download?platform=macArm"
 							className="flex items-center rounded-xl border border-border bg-card/60 backdrop-blur-sm px-5 py-4 transition-all hover:border-accent/30 hover:bg-card group"
 						>
 							<AppleIcon className="h-6 w-6 shrink-0 text-muted-foreground group-hover:text-foreground transition-colors" />
@@ -418,8 +413,7 @@ export default function Home() {
 
 						{/* macOS Intel */}
 						<a
-							href={downloadLinks.macIntel}
-							download
+							href="/download?platform=macIntel"
 							className="flex items-center rounded-xl border border-border bg-card/60 backdrop-blur-sm px-5 py-4 transition-all hover:border-accent/30 hover:bg-card group"
 						>
 							<AppleIcon className="h-6 w-6 shrink-0 text-muted-foreground group-hover:text-foreground transition-colors" />
@@ -431,8 +425,7 @@ export default function Home() {
 
 						{/* Windows */}
 						<a
-							href={downloadLinks.windows}
-							download
+							href="/download?platform=windows"
 							className="flex items-center rounded-xl border border-border bg-card/60 backdrop-blur-sm px-5 py-4 transition-all hover:border-accent/30 hover:bg-card group"
 						>
 							<WindowsIcon className="h-6 w-6 shrink-0 text-muted-foreground group-hover:text-foreground transition-colors" />

landing/src/components/ApiSection.tsx

@@ -29,8 +29,8 @@ const CURL_SNIPPET = `curl -X POST http://127.0.0.1:17493/generate \\
   -H "Content-Type: application/json" \\
   -d '{
     "text": "Welcome to the game, player one.",
-    "profile_id": "morgan-freeman",
-    "engine": "qwen",
+    "profile_id": "b3f1c2d4-5e6f-4a7b-8c9d-0e1f2a3b4c5d",
+    "engine": "qwen_custom_voice",
     "instruct": "warm, slow, cinematic"
   }' \\
   --output line.wav`;

landing/src/components/Footer.tsx

@@ -45,7 +45,7 @@ export function Footer() {
                 </a>
               </li>
               <li>
-                <a href="#download" className="hover:text-foreground transition-colors">
+                <a href="/download" className="hover:text-foreground transition-colors">
                   Download
                 </a>
               </li>

landing/src/components/Navbar.tsx

@@ -66,7 +66,7 @@ export function Navbar() {
             API
           </a>
           <a
-            href="#download"
+            href="/download"
             className="rounded-md px-3 py-1.5 text-sm font-medium text-muted-foreground transition-colors hover:text-foreground"
           >
             Download

landing/src/components/PlatformIcons.tsx

@@ -1,23 +1,29 @@
+import { SiApple, SiLinux } from '@icons-pack/react-simple-icons';
+
+// Official brand icons via Simple Icons (apple/linux). Simple Icons drops
+// Microsoft's mark due to trademark policy, so the Windows 11 flag is
+// inlined from Microsoft's public brand guidance.
+
 export function AppleIcon({ className }: { className?: string }) {
-  return (
-    <svg className={className} viewBox="0 0 24 24" fill="currentColor">
-      <path d="M17.05 20.28c-.98.95-2.05.88-3.08.4-1.09-.5-2.08-.48-3.24 0-1.44.62-2.2.44-3.06-.4C2.79 15.25 3.51 7.59 9.05 7.31c1.35.07 2.29.74 3.08.8 1.18-.24 2.31-.93 3.57-.84 1.51.12 2.65.72 3.4 1.8-3.12 1.87-2.38 5.98.48 7.13-.57 1.5-1.31 2.99-2.54 4.09l.01-.01zM12.03 7.25c-.15-2.23 1.66-4.07 3.74-4.25.29 2.58-2.34 4.5-3.74 4.25z" />
-    </svg>
-  );
+  return <SiApple className={className} color="currentColor" />;
+}
+
+export function LinuxIcon({ className }: { className?: string }) {
+  return <SiLinux className={className} color="currentColor" />;
 }
 
 export function WindowsIcon({ className }: { className?: string }) {
   return (
-    <svg className={className} viewBox="0 0 24 24" fill="currentColor">
-      <path d="M3 12V6.75l6-1.32v6.48L3 12zm17-9v8.75l-10 .15V5.21L20 3zM3 13l6 .09v7.81l-6-1.15V13zm17 .25V22l-10-1.8v-7.15l10 .15z" />
-    </svg>
-  );
-}
-
-export function LinuxIcon({ className }: { className?: string }) {
-  return (
-    <svg className={className} viewBox="0 0 24 24" fill="currentColor">
-      <path d="M12.504 0c-.155 0-.315.008-.48.021-4.226.333-3.105 4.807-3.17 6.298-.076 1.092-.3 1.953-1.05 3.02-.885 1.051-2.127 2.75-2.716 4.521-.278.832-.41 1.684-.287 2.489a.424.424 0 00-.11.135c-.26.26-.195.69-.133 1.001.054.27.112.553.077.784-.12.794-.3 1.593-.3 2.406 0 .599.18 1.193.3 1.791.12.599.3 1.193.3 1.792 0 .812.18 1.611.3 2.405.035.23-.023.514-.077.783-.062.312-.127.742.133 1.002a.424.424 0 00.11.135c-.123.805.01 1.657.287 2.489.589 1.771 1.831 3.47 2.716 4.521.75 1.067 0.974 1.928 1.05 3.02.065 1.491-1.056 5.965 3.17 6.298.165.013.325.021.48.021.155 0 .315-.008.48-.021 4.226-.333 3.105-4.807 3.17-6.298.076-1.092.3-1.953 1.05-3.02.885-1.051 2.127-2.75 2.716-4.521.278-.832.41-1.684.287-2.489a.424.424 0 00.11-.135c.26-.26.195-.69.133-1.001-.054-.27-.112-.553-.077-.784.12-.794.3-1.593.3-2.406 0-.599-.18-1.193-.3-1.791-.12-.599-.3-1.193-.3-1.792 0-.812-.18-1.611-.3-2.405-.035-.23.023-.514.077-.783.062-.312.127-.742-.133-1.002a.424.424 0 00-.11-.135c.123-.805-.01-1.657-.287-2.489-.589-1.771-1.831-3.47-2.716-4.521-.75-1.067-.974-1.928-1.05-3.02-.065-1.491 1.056-5.965-3.17-6.298C12.819.008 12.659 0 12.504 0z" />
+    <svg
+      className={className}
+      viewBox="0 0 24 24"
+      fill="currentColor"
+      xmlns="http://www.w3.org/2000/svg"
+      role="img"
+      aria-label="Windows"
+    >
+      <title>Windows</title>
+      <path d="M0 3.449L9.75 2.1v9.451H0m10.949-9.602L24 0v11.4l-13.051.149M0 12.6h9.75v9.451L0 20.699M10.949 12.6H24V24l-12.9-1.801" />
     </svg>
   );
 }

package.json

@@ -1,6 +1,6 @@
 {
   "name": "voicebox",
-  "version": "0.4.1",
+  "version": "0.4.2",
   "private": true,
   "workspaces": [
     "app",

tauri/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@voicebox/tauri",
   "private": true,
-  "version": "0.4.1",
+  "version": "0.4.2",
   "type": "module",
   "scripts": {
     "dev": "vite",

tauri/src-tauri/Cargo.lock

@@ -5041,7 +5041,7 @@ checksum = "0b928f33d975fc6ad9f86c8f283853ad26bdd5b10b7f1542aa2fa15e2289105a"
 
 [[package]]
 name = "voicebox"
-version = "0.4.1"
+version = "0.4.2"
 dependencies = [
  "base64 0.22.1",
  "core-foundation-sys",

tauri/src-tauri/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "voicebox"
-version = "0.4.1"
+version = "0.4.2"
 description = "A production-quality desktop app for Qwen3-TTS voice cloning and generation"
 authors = ["you"]
 license = ""

tauri/src-tauri/tauri.conf.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://schema.tauri.app/config/2",
   "productName": "Voicebox",
-  "version": "0.4.1",
+  "version": "0.4.2",
   "identifier": "sh.voicebox.app",
   "build": {
     "beforeDevCommand": "bun run dev",

web/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@voicebox/web",
   "private": true,
-  "version": "0.4.1",
+  "version": "0.4.2",
   "type": "module",
   "scripts": {
     "dev": "vite",