# Pocket TTS

> Kyutai's voice model that synthesizes and clones voice on CPU

- Category: Image, video & voice (`media-ai`)
- Tags: `open-source`, `free`
- Official: https://github.com/kyutai-labs/pocket-tts
- Wunderlust: https://wunderlust.xyz/r/pocket-tts
- Image: https://wunderlust.xyz/media/pocket-tts-poster.jpg
- Video preview: https://wunderlust.xyz/media/pocket-tts.webm
- Added: 2026-11-05

## What it is

An open source text-to-speech model from Kyutai with about 100 million parameters, which runs on CPU with no GPU or API.

## What it’s for

Use it to generate or clone voice locally, through a Python library, CLI or HTTP server.

## Benefits

- Streaming with about 200 ms to the first chunk of audio
- Voice cloning from a WAV file and ready-made voices, in seven languages
- Accepts long texts

## Caveats

- MIT license
- According to the repository, it runs at about 6 times real time on a MacBook Air M4; this is the project's claim
- The usage policy prohibits cloning a voice without explicit consent
- The video's comparisons were not tested

## Similar on Wunderlust

- [VoiceStudio](https://wunderlust.xyz/r/voicestudio.md): Voice, cloning, transcription and dubbing
- [Voicebox](https://wunderlust.xyz/r/voicebox.md): A local, open source voice studio: cloning, TTS and dictation
- [LongCat Video](https://wunderlust.xyz/r/longcat-video.md): Meituan's video model for text, image and continuation, with an avatar variant
- [OpenCut](https://wunderlust.xyz/r/opencut.md): Privacy-focused open source video editor, available in the browser and on GitHub
- [Higgsfield CLI](https://wunderlust.xyz/r/higgsfield-cli.md): A CLI to generate images, video, audio and 3D with Higgsfield models from the terminal
- [Kimi K3 in C](https://wunderlust.xyz/r/kimi-k3-in-c.md): A Kimi K3 inference engine in C99, dependency-free, on CPU