Pocket TTS
A lightweight TTS model developed by Kyutai Labs with only 1.9 million parameters. It supports voice cloning and runs on local devices, making it suitable for embedded speech interaction scenarios.
Timeline (last 90 days)
Kyutai Labs' Pocket TTS model has 1.9 million (1.9M) parameters.
The Pocket TTS model claims to support voice cloning.
The new version of Pocket TTS replaces profanity with phonetically similar nonsense words during speech synthesis (e.g., "What the fuck" becomes "What the dee").
The old version of the Pocket TTS model does not sanitize profanity.
The community speculates that Kyutai is shifting Pocket TTS's positioning toward enterprise users and therefore added content filtering by default.
The posting user said the model can generate racially discriminatory content yet is especially sensitive to profanity.
The community currently lacks public benchmark data to evaluate Pocket TTS's naturalness, real-time factor, and speaker similarity.
All Facts (7)
The old version of the Pocket TTS model does not sanitize profanity.
50%UnverifiedThe Pocket TTS model claims to support voice cloning.
50%UnverifiedThe community currently lacks public benchmark data to evaluate Pocket TTS's naturalness, real-time factor, and speaker similarity.
50%UnverifiedKyutai Labs' Pocket TTS model has 1.9 million (1.9M) parameters.
50%UnverifiedThe community speculates that Kyutai is shifting Pocket TTS's positioning toward enterprise users and therefore added content filtering by default.
50%UnverifiedThe posting user said the model can generate racially discriminatory content yet is especially sensitive to profanity.
50%UnverifiedThe new version of Pocket TTS replaces profanity with phonetically similar nonsense words during speech synthesis (e.g., "What the fuck" becomes "What the dee").
50%