Voicebox
Local desktop voice studio that clones voices, generates speech with 8 TTS engines, transcribes with Whisper and adds system-wide dictation.
Voicebox (v0.5.0 release April 2026, changelog to v0.6.0 in Oct 2026, ~59k stars, last commit 7 Oct 2026) is a Tauri desktop app with a Python backend that runs on your machine. It clones voices from a short sample, generates speech in 23 languages with engines such as Qwen3-TTS, Chatterbox and Kokoro, and has a multi-track Stories editor, audio effects, Whisper transcription, a dictation hotkey, a REST API and an MCP server so agents can speak in your voices. Installers for macOS and Windows, Docker, Linux from source. MIT.
The facts
| Question | Answer | Source | Checked |
|---|---|---|---|
| Runs on my machine? | Yes. Runs entirely on your machine; an optional remote mode runs the backend on another machine such as a GPU server | github.com | Checked 5 days ago |
| Code licence | Open source. MIT | github.com | Checked 5 days ago |
| Model weights licence | With conditions. Each TTS engine downloads its own weights under its own licence (Kokoro and Qwen3-TTS VoiceDesign are Apache-2.0); the opt-in OmniVoice engine has CC-BY-NC weights and is labelled non-commercial | github.com | Checked 5 days ago |
| Commercial use | With conditions. MIT code allows commercial use; output from the OmniVoice engine must not be used commercially, and other engines follow their own weight licences | github.com | Checked 5 days ago |
| Paid services needed? | None; speech generation, transcription and the bundled Qwen3 LLM all run locally | github.com | Checked 5 days ago |
| Hardware | Minimum 8 GB RAM, 5 GB free disk and a modern multi-core CPU; 16 GB+ RAM and a CUDA GPU recommended. Also uses Apple Silicon (MLX), ROCm, DirectML and Intel Arc | github.com | Checked 5 days ago |
| Operating systems | macOS 11+ (Apple Silicon and Intel) and Windows 10+ installers; Linux builds from source (no prebuilt binaries yet); Docker image | github.com | Checked 5 days ago |
| Setup effort | Easy on macOS and Windows; download the DMG or MSI and the first generation downloads its model (about 350 MB to 8 GB). Linux needs a source build | github.com | Checked 5 days ago |
Each fact has its own date. Unknown means nobody has published it. We read their documentation; we didn’t install it. Running it yourself costs your hardware, electricity and setup time. Report a correction.
No licence changes since we started watching. Licence changes: follow Voicebox and we email you if its terms change.
What it replaces, job by job
| Paid tool | Jobs | How far it goes |
|---|---|---|
| ElevenLabs $6/month |
| 3 of 4 jobs. Covers local speech generation, voice cloning and Whisper transcription; no hosted conversational voice agent platform, only an MCP tool that lets coding agents speak. |
| Descript $24/month |
| 2 of 3 jobs. Voice cloning and text-to-speech with a multi-track timeline for audio; no transcript-based audio or video editing. |
Stars over time
Building history Not enough history for a 7-day gain. Stars show attention, not quality. Where these numbers come from.