LaunchAI ModelsOctober 20, 2026

Kyutai shrinks Pocket TTS to 1.9M parameters, 103x faster than real time

Read original source →kyutai.org

Kyutai released smaller Pocket TTS variants from its ~100M-parameter original: an 18.2M model matching the original's quality, a 6.9M model, and a 1.9M model running 103x faster than real time on one CPU core. All retain voice cloning.

2 sources

More stories today

Open the live feed