Skip to content
Get packs

Guide · Open source

How to use pocket-tts: install, generate speech and clone voices

Updated · written with AI from the project's own documentation and checked against our catalogue data

Pocket TTS is an open-source text-to-speech tool from Kyutai that turns written text into spoken audio. It is built to run on an ordinary CPU, so you don't need a graphics card or a paid web API. This guide shows you how to install it, generate a wav file from the command line, run a local server, and clone a voice from your own audio sample.

Pocket TTSkyutai-labs/pocket-tts · ★ 10.2k · MIT

Quick start

  1. 1Install the package. Run pip install pocket-tts in a Python 3.10+ environment that has PyTorch 2.5 or newer.
  2. 2Generate a default clip. Run pocket-tts generate to create ./tts_output.wav with the default text and voice.
  3. 3Change the text or voice. Add --text and --voice options to generate speech you choose.
  4. 4Pick a language. Use --language to select a pretrained model, for example italian_24l for Italian.
  5. 5Start the local server. Run pocket-tts serve and open http://localhost:8000 in your browser.
  6. 6Try voice cloning. Pass a wav file to --voice, or convert it once with export-voice for faster reuse.

What you need before you install pocket-tts

The README lists Python 3.10 or newer and PyTorch 2.5 or newer as prerequisites. Pocket TTS does not require the GPU build of PyTorch.

uv is optional, but the README recommends it because it installs dependencies on the fly in an isolated environment.

  • Python 3.10, 3.11, 3.12, 3.13 or 3.14
  • PyTorch 2.5+ (CPU build recommended on Linux)
  • uv (optional, recommended by the README)

Install pocket-tts with pip

If you want to install it manually, run the pip command below. The README also says that on Linux you should see its CPU-only installation section to avoid pulling in the CUDA build of PyTorch.

pip install pocket-tts

Generate your first audio file with the generate command

The generate command writes a wav file and displays some speed statistics. Running it without options uses the default text and default voice, and saves the result to ./tts_output.wav.

Use --voice to change the voice and --text to change the text. If you installed with uv, run the uvx form; if you installed with pip, run the plain command.

  • Change the text with --text
  • Change the voice with --voice
  • Choose a language with --language (default: english)
uvx pocket-tts generate # or if you installed it manually with pip: pocket-tts generate

Pick a voice or a language

The README provides a small catalog of voices, including English, Italian, Spanish, German, Portuguese and French options. You choose a pretrained language model with --language when running generate, export-voice or serve.

Non-English languages also have 24-layer variants that the README describes as higher quality but slower, such as --language italian_24l. The README lists seven languages: English, French, German, Portuguese, Italian, Spanish and Dutch.

  • English voices include alba, anna, azelma, charles and marius
  • Italian voice: giovanni; Spanish voice: lola; German voice: juergen; Portuguese voice: rafael; French voice: estelle
  • Small English models are available with --language english_18m_2026-10, english_7m_2026-10 or english_2m_2026-10

Run a local server with the serve command

The serve command starts a local HTTP server so you can generate audio with requests. The README says the web interface is faster than the command line for trying many voices and prompts, because the model stays in memory between requests.

  • Open http://localhost:8000 to use the web interface
uvx pocket-tts serve # or if you installed it manually with pip: pocket-tts serve

Clone a voice from your own audio file

The --voice option also accepts a plain wav file for voice cloning. The README recommends cleaning the sample first, because the sample's audio quality is reproduced in the output.

Processing an audio file is relatively slow. The export-voice command converts a sample into a safetensors voice embedding, which loads much faster later.

  • Use a wav or mp3 sample as your source
  • Clean the sample before use, as the README suggests
  • Use export-voice to save a reusable voice file

What people use it for

  • Generating narration or voiceovers on a laptop without a cloud service
  • Adding speech generation to a Python project that has no GPU
  • Trying voice cloning with your own wav or mp3 recording
  • Creating speech offline when privacy matters
  • Testing many voices and prompts through the local server

Frequently asked questions

Is pocket-tts free to use?

The project is released under the MIT licence, as listed in its repository. Each voice has its own licence, and the README links a page that details them, so check that page before using a specific voice commercially.

Does pocket-tts need a GPU?

No. The README says Pocket TTS is designed to run on CPUs and does not require the GPU version of PyTorch. It does require PyTorch 2.5 or newer.

How do I install pocket-tts?

Run pip install pocket-tts. The README also recommends uv, which runs pocket-tts with uvx pocket-tts generate without a manual install.

Which languages does pocket-tts support?

The README lists English, French, German, Portuguese, Italian, Spanish and Dutch. It says more languages may be added in the future, so check the README for the current list.

Can pocket-tts clone my voice?

Yes. Pass a plain wav file to the --voice option. The README advises cleaning the sample first, because its audio quality is reproduced in the output.

Is there an alternative if I don't want to install anything?

The README links a website demo at kyutai.org/pocket-tts where you can type text, select voices and generate speech in your browser without installing anything.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Pocket TTS suits Python developers and people who want local, private narration on an ordinary laptop, especially if they want voice cloning. It is a small, CPU-focused model, so if you need the highest possible audio quality from a large GPU model or many languages beyond the listed seven, look elsewhere.

Install options & downloads