# Pocket TTS

> Lightweight, CPU-friendly text-to-speech model with fast voice cloning.

Pocket TTS is an open-source 100M-parameter speech synthesis model developed by Kyutai Labs. It generates high-quality speech faster than real-time on standard laptop CPUs without requiring a GPU. It bridges the gap between massive server-hosted TTS models and tiny fixed-voice models by supporting rapid voice cloning from short audio references.

Website: https://github.com/kyutai-labs/pocket-tts

## What it is for

- Synthesize speech locally on laptops and edge hardware without GPU infrastructure
- Clone speaker voices using a short 5-second audio reference clip
- Integrate fast, streaming text-to-speech into local CLI tools or Python applications

## Features

- Runs faster than real-time on standard CPUs (demonstrated over 3x real-time on Apple Silicon)
- Continuous Audio Language Model (CALM) architecture predicting continuous latents
- Voice cloning capturing tone, acoustic conditions, and accents from short samples
- One-step sampling via Lagrangian Self-Distillation (LSD) and drifting heads
- Simple CLI interface executable directly via uvx
- Multilingual support with checkpoints for English, French, Italian, German, Spanish, and more
- Comes with predefined voice embeddings and evaluation scripts

## Pricing

Open source. Free and open source under the MIT license.

## Details

- Type: AI model
- Category: [AI](https://xiry.app/topics/ai.md)
- Open source: yes
- Platforms: Command line, Code library, macOS, Linux, Windows
- For: Developers, AI researchers, and hobbyists looking for fast, local text-to-speech with voice cloning.
- Alternative to: Kokoro, F5-TTS, Chatterbox Turbo, ElevenLabs
- Tags: tts, text to speech, voice cloning, audio, cpu inference, open source

## Why it was saved

- [@kyutai_labs, 13 January 2026](https://x.com/kyutai_labs/status/2011047335892303875): Bridges the gap between huge GPU-bound LLM TTS models and small models like Kokoro (82M) that lack flexible voice cloning. True voice cloning from just 5 seconds of audio: captures voice color, emotion/accent, and acoustic conditions (reverb, mic quality). Runs via a simple CLI (\`uvx pocket-tts generate --text ... --voice sample.wav\`).

## Compare

- [Alternatives to ElevenLabs](https://xiry.app/alternatives/elevenlabs.md)

## Related

- [Coqui TTS](https://xiry.app/resources/coqui-tts): Open-source deep learning toolkit for text-to-speech synthesis and voice cloning. (GitHub repo, Open source)
- [Fish Audio](https://xiry.app/resources/fish-audio): AI text-to-speech, instant voice cloning, and emotional speech control (Web app, Freemium)
- [Artlist AI Voiceover](https://xiry.app/resources/artlist-ai-voiceover): AI voice generator offering realistic narration, voice cloning, and speech-to-speech tools. (Web app, Paid)
- [Chatterbox](https://xiry.app/resources/chatterbox): Open-source zero-shot text-to-speech and voice cloning model (AI model, Open source)
- [NaturalReader](https://xiry.app/resources/naturalreader): AI text-to-speech platform with multilingual voices and voice cloning. (Web app, Freemium)
- [ElevenLabs](https://xiry.app/resources/elevenlabs): AI voice generator, voice cloning, and conversational agent platform (Web app, Freemium)

Updated 10 October 2026. Page: https://xiry.app/resources/pocket-tts

Make your own library: Xiry takes your saved reels, tweets, Reddit bookmarks and Watch Later videos, makes them searchable and pulls the useful stuff out, like this library. It's invite-only for now; join the waitlist at https://xiry.app/#waitlist.