Silencio-2

About

Silencio-2 is a small English text-to-speech model that produces 24 kHz speech from text and can clone a voice from a few seconds of reference audio. Sound tags such as sigh or laugh and emotion tags such as whisper or angry are written inside the text to direct delivery, and an unknown tag raises an error instead of being read aloud. It is a fine-tune of Kyutai's Pocket TTS 18M, ships as a single 112 MB weights file, and the model card reports about 20x faster than real time on four CPU threads of an Apple M4.