Skip to content
OpenSourceSoftware.ai
Voicebox logo

Voicebox

jamiepine · Audio

GitHub stars

58,993
—Awaiting first sync

Awaiting first sync

No measured star history yet

What is Voicebox?

Voicebox lets you clone voices and generate speech without sending audio to a cloud service. The same app handles dictation, so you can speak into other programs and keep a record of what you said. It is free to use, and the models and recordings stay on your machine.

Voicebox is a desktop voice studio that runs AI speech models locally on your computer. It clones a voice from a short audio sample, turns text into speech with several open models, and transcribes dictation with Whisper. A built-in REST API and MCP server let other apps and AI agents speak and transcribe through it.

Core capabilities

  • Clones voices from audio samples using engines such as Qwen3-TTS, Chatterbox and Kokoro.
  • Dictates with a global hotkey, using Whisper and an optional local clean-up model.
  • Lets MCP-aware agents such as Claude Code and Cursor speak in a chosen voice.

Voice cloning and speech generation

Voice profiles are built from uploaded audio or in-app recordings, and can hold several samples for better results. You can switch between engines for each generation, including Qwen3-TTS, Qwen CustomVoice, LuxTTS, Chatterbox Multilingual, Chatterbox Turbo, HumeAI TADA and Kokoro. Chatterbox Multilingual covers 23 languages, and Chatterbox Turbo reads tags such as [laugh] and [sigh]. Effects such as pitch shift, reverb, delay and compression can be saved as presets. The Stories editor arranges clips from several voices on a multi-track timeline for conversations and podcasts.

  • Voice cloning
  • Text to speech
  • Audio effects
  • Timeline editor

Dictation, captures and agent voices

A global hotkey records speech from anywhere on your computer, and Whisper turns it into text in sizes from Base to Large and Turbo. An optional local Qwen3 model can tidy the transcript before it is used. Each capture keeps the original audio with its transcript, and can be transcribed again, edited or played back in a cloned voice. A REST API and an MCP server give other tools access to speech, transcription, captures and voice profiles. Voice profiles can also carry a written personality that a local model uses to rewrite text in character.

  • Dictation
  • Speech to text
  • MCP
  • REST API

Run it yourself under MIT

Voicebox is a Tauri desktop app with a Python FastAPI backend and a SQLite database. Installers are available for macOS on Apple Silicon and Intel, and for Windows. Linux users build it from source, and Docker Compose runs it as a headless server with a web interface on port 17493. GPU acceleration covers Apple Silicon through MLX, NVIDIA through CUDA, AMD through ROCm, Intel Arc and DirectML, with CPU as a fallback. Voicebox is released under the MIT licence.

  • Desktop app
  • Local AI
  • Docker
  • MIT

Where it fits

Use cases

  1. 01

    Narrate a script in a cloned voice

    Create a voice profile by uploading an audio file or recording a sample in the app, then pick a speech engine and paste your text. Long scripts are split at sentence boundaries and joined with a crossfade, up to 50,000 characters. Apply effects such as reverb or pitch shift, and keep several takes and versions of each generation.

  2. 02

    Dictate into other apps with a hotkey

    Set a push-to-talk or toggle chord in Settings, hold it anywhere, speak and release. Whisper transcribes the audio, and an optional local Qwen3 model removes fillers and false starts. On macOS the text is pasted into the field you were typing in, and every recording and transcript is kept in the Captures tab.

  3. 03

    Hear updates from your coding agents

    Add the Voicebox MCP server at http://127.0.0.1:17493/mcp to Claude Code, Cursor or another MCP client. In Settings, MCP, bind a voice profile to each client so you can tell agents apart. An on-screen pill shows whenever an agent is speaking, and scripts can use the POST /speak endpoint instead.