Voicebox, an open-source AI voice studio by developer Jamie Pine, has launched as a local-first alternative to cloud voice services like ElevenLabs (speech output) and WisprFlow (speech input). The project, built with Tauri and Rust, runs entirely on the user's machine with no cloud dependency.
The app bundles seven distinct TTS engines — Qwen3-TTS, Qwen CustomVoice, LuxTTS, Chatterbox Multilingual, Chatterbox Turbo, HumeAI TADA, and Kokoro — each with different strengths. Users can switch engines per-generation, clone voices from a few seconds of reference audio, and apply post-processing effects like pitch shift, reverb, and compression via Spotify's Pedalboard library.
On the input side, Voicebox provides global dictation: hold a hotkey anywhere on the system, speak, and the transcript pastes into the focused text field. Whisper-based transcription runs locally, and an optional local LLM (Qwen3) can clean up filler words and false starts before pasting.
The project's MCP (Model Context Protocol) integration is a notable feature for developers. A built-in MCP server lets AI coding agents like Claude Code and Cursor speak output through Voicebox in a cloned voice — one tool call and the agent says "Deploy complete" in the user's chosen voice. Per-agent voice bindings let users assign different voices to different AI tools.
Voicebox supports 23 languages, generates speech in unlimited length via auto-chunking with crossfade, and includes a multi-track Stories editor for podcasts and narratives. It runs on macOS (Apple Silicon via MLX), Windows (CUDA), Linux (CUDA/ROCm), and Docker.
The project targets the growing gap between cloud voice APIs — which are powerful but send all audio data to remote servers — and fully local alternatives that have historically lagged in quality and features. Voicebox aims to close that gap by combining multiple best-in-class open-source models into a single polished application.
