What it does
Voicebox runs the voice input and output loop on your own machine. The README says models, voice data and captures never leave the computer. It clones a voice from a few seconds of audio or uses preset voices, generates speech with eight engines (Qwen3-TTS, Qwen CustomVoice, Qwen VoiceDesign, LuxTTS, Chatterbox Multilingual, Chatterbox Turbo, HumeAI TADA and Kokoro) across 23 languages, and applies post-processing effects such as pitch shift, reverb and compression. A stories editor offers a multi-track timeline for conversations, podcasts and narratives. For input, a global hotkey starts dictation using Whisper, and a Captures tab keeps each recording with its transcript so it can be re-transcribed or cleaned up by a bundled local Qwen3 model. The same local model can rewrite text in a voice profile's free-form personality before it is spoken. A REST API and a built-in MCP server let tools such as Claude Code, Cursor and Cline speak in a cloned voice through a single tool call. Desktop builds are listed for macOS (Apple Silicon and Intel) and Windows, with Docker available and Linux built from source, using GPU acceleration where available.
Capabilities
- Voice cloning from a short reference sample, plus 50+ preset voices
- Eight speech engines covering 23 languages, with expressive tags on one engine
- Global dictation hotkey with Whisper-based transcription and a Captures history
- Bundled local LLM for personalities and transcript cleanup
- REST API and MCP server so agents can speak in a chosen voice
- Multi-track stories editor and audio effects
When to use it
- Generating narration or character voices without sending audio to a cloud service
- Dictating into any text field with a hotkey
- Giving a coding agent a spoken voice through MCP
- Producing multi-voice conversations or podcast-style audio on your own machine
