Voicebox
voicebox.sh- Category
- AI Tools
- Rank
- No. 177Tools index
- Pricing
- Open Source
- Platform
- desktop
- Type
- APP
- Builder
- @jamiepine
- GitHub
- 54.3k stars
- Latest release
- v0.5.0
- Date
About
Open source voice cloning desktop application that lets you clone any voice from just 3 seconds of audio and generate natural speech locally. Features multi-voice project composition, audio effects, and support for multiple TTS engines with complete privacy.
What it does
Voicebox combines speech generation, transcription, system-wide dictation, and agent speech in one local studio. Users choose among seven speech engines, manage cloned or preset voices, edit multi-speaker timelines, and refine text with a bundled local language model. REST and MCP interfaces expose the same voice workflows to other software.
Why it's ranked here
The breadth is unusually practical: one application covers spoken input, generated output, editing, effects, and agent integration while keeping current processing on the user’s machine. The verdict is mixed because the project status also documents serious Apple Silicon regressions, MCP compatibility problems, and an unresolved local-server exposure report.
What's good
Engine choice is meaningful rather than cosmetic. Different options target multilingual cloning, low-memory CPU use, preset speakers, long-form coherence, or expressive tags. Long scripts get sentence-aware chunking and crossfades. Generations retain originals, alternate takes, effects versions, and lineage. A serial asynchronous queue avoids competing jobs for GPU memory.
Tradeoffs
Capabilities vary sharply by engine. Expressive tags work only with Chatterbox Turbo, while several alternatives speak those tags literally. Linux lacks prebuilt desktop binaries and requires a source build. Automatic dictation paste is documented specifically for macOS. Current project notes report model-loading crashes on Apple Silicon, broken paste behavior, scrambled MCP audio, and dotted MCP tool names rejected by some clients.
How to use it well
It best fits creators and developers who want local voice production, reusable profiles, long-form narration, dictation, or spoken output from MCP-aware agents. Match the engine to language, hardware, and expression needs, then preserve clean originals before applying effects. Treat it as voice infrastructure, not a replacement for full agent orchestration or real-time translation.
Technical notes+
backend/app.py builds a FastAPI application, registers routers, mounts a FastMCP Streamable HTTP application at /mcp, configures local CORS origins, serves an optional React SPA, initializes SQLite-backed services, and detects CUDA, ROCm, MPS, MLX, and Intel XPU paths. package.json defines a Bun workspace spanning app, tauri, web, and landing, with Tauri release builds plus TypeScript and Biome checks. requirements.txt establishes the Python server base around FastAPI, Uvicorn, SQLAlchemy, PyTorch, SoundFile, Librosa, multipart uploads, and Hugging Face model access. docs/PROJECT_STATUS.md describes seven concrete TTS backends behind a shared protocol and a serial generation flow. Some planning documents contain stale pre-shipping states, so shipped behavior is better supported by the README, project status, MCP plan, and application code.
Observed
- Implementation stack
- Tauri and Rust desktop shell, React and TypeScript frontend, Python FastAPI backend, and SQLite persistence.
- Packaging and install surface
- DMG packages are listed for Apple Silicon and Intel macOS, MSI for Windows, Docker Compose for containers, and source-build instructions for Linux.
- Interfaces
- Desktop and web interfaces, REST API, built-in MCP server over Streamable HTTP, and a bundled stdio MCP shim.
- Platform support
- The repository describes macOS through MLX or Metal, Windows through CUDA, Linux, AMD ROCm, Intel Arc, CPU execution, and Docker.
- Workspace structure
- The Bun workspace contains four packages: app, tauri, web, and landing.
- Backend architecture
- Seven TTS engine implementations share a protocol and thread-safe engine registry; Whisper provides speech recognition.
Read from README.md, package.json, requirements.txt, docs/README.md, docs/PROJECT_STATUS.md, docs/plans/VOICE_IO.md, docs/plans/MCP_SERVER.md, docs/plans/CLOUD_ROADMAP.md, docs/plans/OPENAI_SUPPORT.md, docs/plans/DOCKER_DEPLOYMENT.md, docs/plans/MACOS_NOTARIZATION.md, docs/content/docs/README.md, backend/app.py.
What it can do
Clone voice from audio sample
3 seconds of audio recording → Voice model for speech synthesis
Generate natural speech
Text and cloned voice model → Synthesized audio file
Compose multi-voice projects
Text scripts and multiple voice models → Audio composition with different voices
Apply audio effects to generated speech
Synthesized audio and effect parameters → Processed audio with applied effects
Process text-to-speech locally
Text content → Audio speech without cloud processing
Tags
Tech Stack
Media
Featured in Intel
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.