Pocket TTS is a lightweight text-to-speech system from Kyutai that runs efficiently on plain CPUs — no GPU, no cloud API, just a pip install and a function call. True to its name, the model is only about 100 million parameters, yet it delivers high-quality speech with voice cloning. The open-source project is approaching 10,000 stars on GitHub and was actively updated through October 2026.
Kyutai designed Pocket TTS for the many situations where GPU servers are impractical: desktop apps, edge devices, accessibility tools, and local assistants. It requires PyTorch 2.5+ and supports Python 3.10 through 3.14. You can pick from built-in voices or clone a voice from a short audio sample, and generate audio through the Python API, a command-line interface, or a local web server. A browser demo on the Kyutai site lets you try it instantly. Language support currently covers English, French, German, Portuguese, Italian, Spanish, and Dutch, with more planned. Voices can be selected per call, generation can stream for responsive playback, and because everything runs on CPU the same code behaves the same on a laptop, a Raspberry Pi-class device, or a small server.
Free. Pocket TTS is open source under the MIT license; the model and code are free to download and use, including in your own products.
Developers building offline or privacy-sensitive speech features — screen readers, local voice assistants, e-learning tools, and embedded apps — who need good TTS without GPU infrastructure or per-character API bills.