Voicebox
Whisper dictation plus local text-to-speech and voice cloning, all on-device.
What it does
Voicebox is an open-source voice studio for macOS that covers both directions of speech:
- Dictation — hold a hotkey anywhere, speak, release, and the transcript is pasted into the focused field. Transcription is OpenAI Whisper running locally, with optional cleanup by a local LLM.
- Synthesis and cloning — generate speech through a choice of local TTS engines, and clone a voice from a sample to reuse it.
- Captures — a paired archive of audio plus transcript, so a recording can be re-transcribed or refined later.
- MCP — it exposes itself to MCP-aware agents, so an agent can speak through one of your voices.
Notes
Not in Homebrew; download from the project site or build from the repository. Models are downloaded on first use and are large.
Voice cloning has real misuse potential. Clone voices you own or have explicit permission to use, and be aware that synthetic speech impersonating a real person is illegal in a growing number of jurisdictions.
Alternative to
| Alternative | Type | Trade-off |
|---|---|---|
| ElevenLabs | Freemium | The commercial standard for TTS and voice cloning; cloud only |
| FluidVoice or Superwhisper | Open source / Freemium | Dictation only, without the synthesis half |
| Piper or Kokoro on their own | Open source | Local TTS engines without the surrounding app |