Community · Nov 1, 2026, 18:07 UTC

Silencio-2, an 18M open-weights TTS with emotion and sound tags

A fine-tune of Kyutai's Pocket TTS that follows inline tags like [sigh], <whisper> and <angry>. It runs about 20x realtime on an Apple M4 CPU, supports voice cloning from a short clip, and is English only. The author did not run a formal WER benchmark on the released checkpoint.

• LabKyutai
• ModelSilencio-2
• ModelPocket TTS
• Topicaudio
• Topicopen-weights
• Topicresearch