voicebox

Voicebox is an open-source AI voice studio designed to run locally on your own machine, positioning itself as a free alternative to cloud services like ElevenLabs and WisprFlow. It covers both voice output (text-to-speech and voice cloning) and voice input (speech-to-text dictation), with a bundled local LLM for optional refinement and per-profile voice personalities. Key capabilities described in the repository: - **Voice cloning and TTS**: Zero-shot cloning from a short reference sample, plus 50+ curated preset voices. Seven TTS engines are included (Qwen3-TTS, Qwen CustomVoice, LuxTTS, Chatterbox Multilingual, Chatterbox Turbo, HumeAI TADA, and Kokoro), covering up to 23 languages depending on engine. - **Speech-to-text**: Powered by OpenAI Whisper locally, with model size options from Base to Turbo, running via MLX on Apple Silicon or PyTorch on CUDA/ROCm/DirectML/CPU. - **Global dictation**: A system-wide hotkey lets users dictate into any focused text field. The README notes target-aware paste with accessibility-verified injection on macOS and first-run permissions UX for Accessibility and Input Monitoring. - **Audio post-processing**: Eight effects built on Spotify's pedalboard library, including pitch shift, reverb, delay, chorus, compressor, gain, and filters, with reusable presets. - **Agent voice output**: A built-in MCP (Model Context Protocol) server exposes tools such as `voicebox.speak`, `voicebox.transcribe`, `voicebox.list_captures`, and `voicebox.list_profiles`. MCP-aware agents like Claude Code, Cursor, Windsurf, and Cline can speak to the user in a cloned voice through one tool call. - **Local privacy**: Models, voice data, and captures stay on the user's machine. - **Stories editor**: A multi-track timeline for conversations, podcasts, and narratives, with drag-and-drop composition and synchronized playback. - **Captures**: Dictations, in-app recordings, and uploads are kept with audio plus transcript, and can be re-transcribed, refined, edited, or promoted to voice samples. - **API**: A REST API is available at `http://127.0.0.1:17493` for `/generate`, `/speak`, `/transcribe`, and `/profiles`, with docs at `/docs`. - **Platforms**: Ready downloads for macOS (Apple Silicon and Intel) and Windows, plus Docker support and build-from-source instructions for Linux. Technically, the desktop app is built with Tauri (Rust) and React/TypeScript, with a FastAPI Python backend, SQLite storage, and MLX or PyTorch inference depending on platform and GPU. The roadmap in the README mentions future work on Windows/Linux auto-paste, additional STT engines, streaming transcription, end-to-end speech LLMs, and plugin architecture. The project is available on GitHub under a license badge noted in the README, and contribution guidelines are included in the repository.