Published on August 25, 2026, the stack includes a data pipeline, training recipes, and evaluations for a text-to-speech model designed to run on CPU.
Kyutai has open-sourced the Pocket TTS training stack. It includes the data pipeline, training recipes, and evaluations; the project was developed to train a text-to-speech model and run it on CPU.
Engineers can examine the complete workflow, from data preparation to evaluation, and inspect or adapt its components. To verify the details, consult Kyutai’s original repository and documentation, checking the published contents and available evaluations.