Pocket TTS is a lightweight text-to-speech tool from Kyutai Labs, built to run entirely on a regular CPU instead of needing a GPU or a cloud API. It’s a tiny 100-million-parameter model, yet it starts producing audio in around 200ms and generates speech roughly six times faster than real-time on something like a MacBook Air M4, all while using just two CPU cores. It handles text of any length, currently supports six languages (English, French, German, Portuguese, Italian, and Spanish), and comes with a library of ready-made voices plus voice cloning from your own audio samples.
Kyutai offers a free browser demo with no install needed, and the actual model is entirely open-source under the MIT license, so for local installs you can check out the GitHub repo.
