SYLIN

Pocket TTS

by Kyutai · Model · Voice & audio

Kyutai released the full training stack: data pipeline, training recipes and evaluation scripts, to train Pocket TTS in any language or voice

AccessOpen weightsLicenseMIT

Evidence

  1. Aug 30
    Four stages: prepare data, forced alignment, train a 24-layer teacher, distill into a 6-layer student that runs on CPU
  2. Aug 21
    Voice-assistant builder uses Pocket TTS for being super fast on-device; wants such quality on phones, not H100s

Alternatives

Built on Pocket TTS