v0.3.0

This release rewrites the backend into a modular architecture, overhauls the settings UI into routed sub-pages, fixes audio player freezing, migrates documentation to Fumadocs, and ships a batch of bug fixes targeting the most-reported issues from the tracker.

The backend's 3,000-line monolith main.py has been decomposed into domain routers, a services layer, and a proper database package. A style guide and ruff configuration now enforce consistency. On the frontend, settings have been split into dedicated routed pages with server logs, a changelog viewer, and an about page. The audio player no longer freezes mid-playback, and model loading status is now visible in the UI. Seven user-reported bugs have been fixed, including server crashes during sample uploads, generation list staleness, cryptic error messages, and CUDA support for RTX 50-series GPUs.

Settings Overhaul (#294)

• Split settings into routed sub-tabs: General, Generation, GPU, Logs, Changelog, About
• Added live server log viewer with auto-scroll
• Added in-app changelog page that parses CHANGELOG.md at build time
• Added About page with version info, license, and generation folder quick-open
• Extracted reusable SettingRow component for consistent setting layouts

Audio Player Fix (#293)

• Fixed audio player freezing during playback
• Improved playback UX with better state management and listener cleanup
• Fixed restart race condition during regeneration
• Added stable keys for audio element re-rendering
• Improved accessibility across player controls

Backend Refactor (#285)

• Extracted all routes from main.py into 13 domain routers under backend/routes/ — main.py dropped from ~3,100 lines to ~10
• Moved CRUD and service modules into backend/services/, platform detection into backend/utils/
• Split monolithic database.py into a database/ package with separate models, session, migrations, and seed modules
• Added backend/STYLE_GUIDE.md and pyproject.toml with ruff linting config
• Removed dead code: unused _get_cuda_dll_excludes, stale studio.py, example_usage.py, old Makefile
• Deduplicated shared logic across TTS backends into backends/base.py
• Improved startup logging with version, platform, data directory, and database stats
• Fixed startup database session leak — sessions now rollback and close in finally block
• Isolated shutdown unload calls so one backend failure doesn't block the others
• Handled null duration in story_items migration
• Reject model migration when target is a subdirectory of source cache

Documentation Rewrite (#288)

• Migrated docs site from Mintlify to Fumadocs (Next.js-based)
• Rewrote introduction and root page with content from README
• Added "Edit on GitHub" links and last-updated timestamps on all pages
• Generated OpenAPI spec and auto-generated API reference pages
• Removed stale planning docs (CUDA_BACKEND_SWAP, EXTERNAL_PROVIDERS, MLX_AUDIO, TTS_PROVIDER_ARCHITECTURE, etc.)
• Sidebar groups now expand by default; root redirects to /docs
• Added OG image metadata and /og preview page

UI & Frontend

• Added model loading status indicator and effects preset dropdown (3187344)
• Fixed take-label race condition during regeneration
• Added accessible focus styling to select component
• Softened select focus indicator opacity
• Addressed 4 critical and 12 major issues from CodeRabbit review

Bug Fixes (#295)

• Fixed sample uploads crashing the server — audio decoding now runs in a thread pool instead of blocking the async event loop (#278)
• Fixed generation list not updating when a generation completes — switched to refetchQueries for reliable cache busting, added SSE error fallback, and page reset on completion (#231)
• Fixed error toasts showing [object Object] instead of the actual error message (#290)
• Added Whisper model selection (base, small, medium, large, turbo) and expanded language support to the /transcribe endpoint (#233)
• Upgraded CUDA backend build from cu121 to cu126 for RTX 50-series (Blackwell) GPU support (#289)
• Handled client disconnects in SSE and streaming endpoints to suppress [Errno 32] Broken Pipe errors (#248)
• Fixed Docker build failure from pip hash mismatch on Qwen3-TTS dependencies (#286)
• Added 50 MB upload size limit with chunked reads to prevent unbounded memory allocation on sample uploads
• Eliminated redundant double audio decode in sample processing pipeline

Platform Fixes

• Replaced netstat with TcpStream + PowerShell for Windows port detection (#277)
• Fixed Docker frontend build and cleaned up Docker docs
• Fixed macOS download links to use .dmg instead of .app.tar.gz
• Added dynamic download redirect routes to landing site

Release Tooling

• Added draft-release-notes and release-bump agent skills
• Wired CI release workflow to extract notes from CHANGELOG.md for GitHub Releases
• Backfilled changelog with all historical releases

v0.2.3

The "it works in dev but not in prod" release. This version fixes a series of PyInstaller bundling issues that prevented model downloading, loading, generation, and progress tracking from working in production builds.

Model Downloads Now Actually Work

The v0.2.1/v0.2.2 builds could not download or load models that weren't already cached from a dev install. This release fixes the entire chain:

• Chatterbox, Chatterbox Turbo, and LuxTTS all download, load, and generate correctly in bundled builds
• Real-time download progress — byte-level progress bars now work in production. The root cause: huggingface_hub silently disables tqdm progress bars based on logger level, which prevented our progress tracker from receiving byte updates. We now force-enable the internal counter regardless.
• Fixed Python 3.12.0 code.replace() bug — the macOS build was on Python 3.12.0, which has a known CPython bug that corrupts bytecode when PyInstaller rewrites code objects. This caused NameError: name 'obj' is not defined crashes during scipy/torch imports. Upgraded to Python 3.12.13.

PyInstaller Fixes

• Collect all inflect files — typeguard's @typechecked decorator calls inspect.getsource() at import time, which needs .py source files, not just bytecode. Fixes LuxTTS "could not get source code" error.
• Collect all perth files — bundles the pretrained watermark model (hparams.yaml, .pth.tar) needed by Chatterbox at runtime
• Collect all piper_phonemize files — bundles espeak-ng-data/ (phoneme tables, language dicts) needed by LuxTTS for text-to-phoneme conversion
• Set ESPEAK_DATA_PATH in frozen builds so the espeak-ng C library finds the bundled data instead of looking at /usr/share/espeak-ng-data/
• Collect all linacodec files — fixes inspect.getsource error in Vocos codec
• Collect all zipvoice files — fixes source code lookup in LuxTTS voice cloning
• Copy metadata for requests, transformers, huggingface-hub, tokenizers, safetensors, tqdm — fixes importlib.metadata lookups in frozen binary
• Add hidden imports for chatterbox, chatterbox_turbo, luxtts, zipvoice backends
• Add multiprocessing.freeze_support() to fix resource_tracker subprocess crash in frozen binary
• --noconsole now only applied on Windows — macOS/Linux need stdout/stderr for Tauri sidecar log capture
• Hardened sys.stdout/sys.stderr devnull redirect to test writability, not just None check

Updater

• Fixed updater artifact generation with v1Compatible for tauri-action signature files
• Updated tauri-action to v0.6 to fix updater JSON and .sig generation

Other Fixes

• Full traceback logging on all backend model loading errors (was just str(e) before)

v0.2.2

UPDATE: I'm working on a rewrite of the model downloading, it's absolute hell and takes a while to test as it always works in dev and never in prod builds. Will have a solution up ASAP. If you're eager to test 0.2.x please compile from source. Next update will solve model downloading and the updater issue for good.

v0.2.2

• Fix Chatterbox model support in bundled builds [SIKE fixed in 0.2.3]
• Fix LuxTTS/ZipVoice support in bundled builds [SIKE fixed in 0.2.3]
• Auto-update CUDA binary when app version changes
• CUDA download progress bar
• Fix server process staying alive on macOS (SIGHUP handling, watchdog grace period)
• Hide console window when running CUDA binary on Windows

v0.2.1

The best local voice cloning tool, just got better...

See the new website: https://voicebox.sh

Released 2026-03-15 — v0.2.1 on GitHub (version bump due to an immutable release tag on GitHub)

Voicebox v0.1.x was a single-engine voice cloning app built around Qwen3-TTS. v0.2.0 is a ground-up rethink: four TTS engines, 23 languages, paralinguistic emotion controls, a post-processing effects pipeline, unlimited generation length, an async generation queue, and support for every major GPU vendor. Plus Docker.

---

New TTS Engines

Multi-Engine Architecture

Voicebox now runs four independent TTS engines behind a thread-safe per-engine backend registry. Switch engines per-generation from a single dropdown — no restart required.

Engine	Languages	Size	Key Strengths
Qwen3-TTS 1.7B	10	~3.5 GB	Highest quality, delivery instructions ("speak slowly", "whisper")
Qwen3-TTS 0.6B	10	~1.2 GB	Lighter, faster variant
LuxTTS	English	~300 MB	CPU-friendly, 48 kHz output, 150x realtime
Chatterbox Multilingual	23	~3.2 GB	Broadest language coverage, zero-shot cloning
Chatterbox Turbo	English	~1.5 GB	350M params, low latency, paralinguistic tags

Chatterbox Multilingual — 23 Languages (#257)

Zero-shot voice cloning in Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, and Turkish. The language dropdown dynamically filters to show only languages supported by the selected engine.

LuxTTS — Lightweight English TTS (#254)

A fast, CPU-friendly English engine. ~300 MB download, 48 kHz output, runs at 150x realtime on CPU. Good for quick drafts and machines without a GPU.

Chatterbox Turbo — Expressive English (#258)

A fast 350M-parameter English model with inline paralinguistic tags.

Paralinguistic Tags Autocomplete (#265)

Type / in the text input with Chatterbox Turbo selected to open an autocomplete for 9 expressive tags that the model synthesizes inline with speech:

[laugh] [chuckle] [gasp] [cough] [sigh] [groan] [sniff] [shush] [clear throat]

Tags render as inline badges in a rich text editor and serialize cleanly to the API.

---

Generation

Unlimited Generation Length — Auto-Chunking (#266)

Long text is now automatically split at sentence boundaries, generated per-chunk, and crossfaded back together. Engine-agnostic — works with all four engines.

• Auto-chunking limit slider — 100–5,000 chars (default 800)
• Crossfade slider — 0–200ms (default 50ms), or 0 for a hard cut
• Max text length raised to 50,000 characters
• Smart splitting respects abbreviations (Dr., e.g., a.m.), CJK punctuation, and never breaks inside [tags]

Asynchronous Generation Queue (#269)

Generation is now fully non-blocking. Submit a generation and start typing the next one immediately.

• Serial execution queue prevents GPU contention
• Real-time SSE status streaming (generating → completed / failed)
• Failed generations can be retried without re-entering text
• Stale generations from crashes are auto-recovered on startup
• Generating status pill shown inline in the story editor

Generation Versions

Every generation now supports multiple versions with provenance tracking:

• Original — the unprocessed TTS output, always preserved
• Effects versions — apply different effects chains to create new versions from any source
• Takes — regenerate with the same text/voice but a new seed
• Source tracking — each version records which version it was derived from
• Version pinning in stories — pin a specific version to a story track clip
• Favorites — star generations for quick access

Language Parameter Fix

Qwen TTS models now correctly receive the selected language. The generation form syncs with the voice profile's language setting.

---

Post-Processing Effects (#271)

A full audio effects system powered by Spotify's pedalboard library. Apply effects after generation, preview in real time, and build reusable presets.

Effect	Description
Pitch Shift	±12 semitones
Reverb	Room size, damping, wet/dry mix
Delay	Adjustable time, feedback, mix
Chorus / Flanger	Modulated delay — short for metallic, long for lush
Compressor	Threshold, ratio, attack, release
Gain	-40 to +40 dB
High-Pass Filter	Configurable cutoff frequency
Low-Pass Filter	Configurable cutoff frequency

• 4 built-in presets — Robotic, Radio, Echo Chamber, Deep Voice
• Custom presets — create unlimited drag-and-drop effect chains
• Per-profile default effects — assign a chain to a voice profile, auto-applies to every generation
• Live preview — audition effects against existing audio before committing
• Source version selection — apply effects to any version of a generation, not just the latest

---

Platform Support

Windows Support (#272)

Full Windows support with CUDA GPU detection, cross-platform justfile, and clean server shutdown using taskkill /T for the process tree.

Linux (#262)

Pre-built Linux binaries are not available for this release — the release CI is still broken on Linux and we're working on fixing it. However, this release includes significant Linux improvements that make compiling from source much easier:

• AMD ROCm GPU acceleration with automatic HSA_OVERRIDE_GFX_VERSION for unlisted GPUs
• NVIDIA GBM buffer crash fix (#210)
• WebKitGTK microphone access for voice sample recording
• Cross-platform justfile with Linux-specific setup targets
• See the README for build-from-source instructions — we'll ship Linux CI builds as soon as we can

NVIDIA CUDA Backend Swap (#252)

The CPU-only release can download and swap in a CUDA-accelerated backend from within the app. Downloads split parts to work around GitHub's 2GB asset limit, verifies SHA-256 checksums, and restarts the server automatically.

Intel Arc (XPU) and DirectML

PyTorch backend supports Intel Arc GPUs via IPEX/XPU and any-GPU on Windows via DirectML.

Docker + Web Deployment (#161)

Run Voicebox headless:

docker compose up

3-stage build, non-root runtime, health checks, persistent model cache. Binds to localhost only by default.

Whisper Turbo

Added openai/whisper-large-v3-turbo as a transcription model option.

---

Model Management (#268)

• Per-model unload — free GPU memory without deleting downloaded models
• Custom models directory — set VOICEBOX_MODELS_DIR to store models anywhere
• Model folder migration — move all models to a new location with progress tracking
• Download cancel/clear UI — cancel in-progress downloads, VS Code-style problems panel for errors (#238)
• Restructured settings UI — server settings and model management split into cleaner sections

---

Security & Reliability

• CORS hardening — explicit allowlist of local origins instead of wildcard *; extensible via VOICEBOX_CORS_ORIGINS (#88)
• Network access toggle — fully disable outbound requests for air-gapped deployments (#133)
• Offline crash fix — Voicebox no longer crashes when HuggingFace is unreachable (#152)
• Atomic audio saves — two-phase write prevents corrupted files on crash or disk-full (#263)
• Filesystem health endpoint — proactive disk space and directory writability checks
• Errno-specific error messages — clear feedback for permission denied, disk full, missing directory
• Chatterbox float64 dtype fix — patches S3Tokenizer and VoiceEncoder to cast float64→float32, preventing crashes on certain audio inputs (#264)
• Watchdog respects keep-server-running — /watchdog/disable endpoint prevents the server from shutting down when the app window closes, if configured
• Server shutdown on Windows — clean process tree termination with taskkill /T and os._exit fallback

---

Accessibility (#243)

• Screen reader support (tested with NVDA/Narrator) across all major UI surfaces
• Keyboard navigation for voice cards, history rows, model management, and story editor
• State-aware aria-label attributes on all interactive controls

---

UI Polish

• Redesigned landing page with animated ControlUI hero, multi-engine copy, model cards, and voice creator section (#274)
• Glassmorphic active state for sidebar buttons with accent border shine
• Voices tab overhaul with inline inspector
• Re...

v0.1.13

What's Changed

Stability and reliability

• #95 Fix: selecting 0.6B model still downloads and uses 1.7B
• #93 fix(mlx): bundle native libs and broaden error handling for Apple Silicon
• #79 fix: handle non-ASCII filenames in Content-Disposition headers
• #78 fix: guard getUserMedia call against undefined mediaDevices in non-secure contexts
• #77 fix: await for confirmation before deleting voices and channels
• #128 fix: resolve multiple issues (#96, #119, #111, #108, #121, #125, #127)
• #40 Fix: audio export path resolution

Build and packaging

• #122 fix(web): add @tailwindcss/vite plugin to web config
• #126 Create requirements.txt

UX and docs

• #44 Enhances floating generate box UX
• #57 chore: updates repo URL in README
• #146 Add Spacebot banner to landing page
• #1 Improvements

v0.1.12

Model Download UX Overhaul

• Real-time download progress tracking with accurate percentage and speed info
• No more downloading notifications during generation even when its not downloading
• Better error handling and status reporting throughout the download process

Other Improvements

• Enhanced health check endpoint with GPU type information
• Improved model caching verification
• More reliable SSE progress updates
• Actual update notifications, you don't need to go to settings and manually check anymore

Note: CUDA support for windows coming next update see issue and my plan.

v0.1.11

• Fixed transcriptions on MLX
• Fixed model download progress (finally)

v0.1.10

Faster generation on Apple Silicon

Massive speed gains, from around 20s per generation to 2-3s!

Added native MLX backend support for Apple Silicon, providing significantly faster TTS and STT generation on M-series macOS machines.

Note: this update broke transcriptions on Apple Silicon only, the patch is in the oven as we speak, 0.1.11 will follow.

Features

• MLX Backend: New backend implementation optimized for Apple Silicon using MLX framework
• Dynamic Backend Selection: Automatically detects platform and selects between MLX (macOS) and PyTorch (other platforms)
• Improved Performance: Leverages Apple's unified memory architecture for faster model inference

Backend Changes

• Refactored TTS and STT logic into modular backend implementations (mlx_backend.py, pytorch_backend.py)
• Added platform detection system to handle backend selection at runtime
• Updated model loading and caching to support both backend types
• Enhanced health check endpoints to report active backend type

Build & Release

• Updated build process to include MLX-specific dependencies for macOS builds
• Modified release workflow to handle platform-specific backend bundling
• Added requirements-mlx.txt for MLX dependencies

Documentation

• Updated setup and building guides with MLX-specific instructions
• Added troubleshooting guidance for MLX-related issues
• Enhanced architecture documentation to explain backend selection

v0.1.9

Improved voice profile creation flow,

• Voice create drafts: No longer lose work if you close the model
• Fixed whisper only transcribing English or Chinese, now has support for all languages

Improved Stories editor:

• Added spacebar for play/pause
• Timeline now auto-scrolls to follow playhead during playback
• Fixed misalignment of the items with mouse when picking up
• Fixed hitbox for selecting an item
• Fixed playhead jumping forward when pressing play (the timing anchors bug)

Generation box improvements

• Instruct mode no longer wipes prompt text
• Improved UI cleanliness

Misc

• Fixed "Model downloading" toast during generation when model is already downloaded

v0.1.8

🐛 Bug Fixes

Model Download Timeout Issues

Fixed critical issue where model downloads would fail with "Failed to fetch" errors on Windows:

• Root Cause: Multi-GB model downloads exceeded HTTP request timeout (30-60s), causing frontend to show errors even though downloads were continuing in background
• Solution: Refactored download endpoints to return immediately and continue downloads in background
• /models/download endpoint now returns instantly with download starting in background
• /generate and /transcribe endpoints now auto-start model downloads when needed
• Returns 202 Accepted status with download progress information for better UX
• Frontend can track download progress via SSE endpoint and retry when complete

Cross-Platform Cache Path Issues

• Fixed hardcoded ~/.cache/huggingface/hub paths that don't work on Windows
• All cache paths now use hf_constants.HF_HUB_CACHE for proper cross-platform support
• Windows: Uses %USERPROFILE%\.cache\huggingface\hub or %LOCALAPPDATA%
• macOS/Linux: Uses ~/.cache/huggingface/hub
• Ensures HuggingFace cache directory exists on startup (defensive fix)

✨ Features

Windows Process Management

• Added /shutdown endpoint for graceful server shutdown on Windows
• Improved process lifecycle management for bundled server binary

GPU Detection Improvements

• Added gpu_type field to health check response
• Now shows specific GPU type: "CUDA (GPU Name)", "MPS (Apple Silicon)", or None
• Fixes UI showing "GPU: Not Available" when MPS/CUDA is actually detected