Radar ·
Kyutai releases smaller Pocket TTS versions with up to 1.9M parameters
Kyutai Labs says it released smaller Pocket TTS versions at 18.2M, 6.9M and 1.9M parameters, with voice cloning and CPU execution. The source reports 103x real-time speed for the 1.9M model on a single core.
According to Kyutai Labs, Pocket TTS, a text-to-speech model of about 100M parameters, now has smaller versions at 18.2M, 6.9M and 1.9M parameters. The source attributes to the 1.9M model a speed of 103 times real time on a single CPU core, with voice cloning. This figure is the author's claim and has not been verified here. The source describes the release as small, self-hostable TTS useful for on-device voice apps.
To consult the original, find the announcement or repository named by Kyutai Labs and check the license, requirements and benchmark methodology before adopting it. Rota Nacional does not offer this model and does not guarantee equivalent performance. If you use AI to study this material, do not send recordings or personal texts of third parties, since cloning a real person's voice requires consent; prefer synthetic or test samples.
Original source: Kyutai lança versões menores do Pocket TTS com até 1,9M de parâmetros