Kyutai Pocket TTS
Gated on HF (accept prohibited-use terms). Training code released 2026-08-25; 2026-09-28 post describes a 'drifting' objective replacing flow matching for the sampler head. `pip install pocket-tts`. Community WebAssembly ports run in-browser.
- Input
- text, audio
- Output
- audio
- License
- cc-by-4.0
- Verified
- 2026-09-29
How to call it
| Provider | Model id | Endpoint / URL | Docs |
|---|---|---|---|
| Hugging Face | kyutai/pocket-tts | huggingface.co/kyutai/pocket-tts | — |
| Hugging Face (no cloning variant) | kyutai/pocket-tts-without-voice-cloning | huggingface.co/kyutai | — |
| GitHub / pip | pocket-tts | github.com/kyutai-labs/pocket-tts | — |
Notable capabilities (2)
- 100M-param TTS with cloning, real time on CPU: ~200 ms to first audio and ~6x real time on a MacBook Air M4 CPU; streaming, unbounded text length; voice cloning from audio. source
- Six languages (found after launch): English, French, German, Spanish, Portuguese, Italian (multilingual since 2026-05-04). source
Sources: https://huggingface.co/kyutai/pocket-tts , https://github.com/kyutai-labs/pocket-tts , https://kyutai.org/blog/
Other Kyutai models
Kyutai TTS 1.6B / Kyutai STT + Unmute · Kyutai Moshi / Hibiki-Zero (full-duplex speech models)