Show HN: Silencio 2 – an 18M TTS with emotion tags that runs on a laptop

Category: ai-ml

Tags: text-to-speech, voice-cloning, emotion-control, small-model, cpu-inference

Score: 6.3/10 (Innovation: 6, Technical: 7, Documentation: 7, Utility: 6)

Silencio-2 is an 18M-parameter English text-to-speech model that supports inline emotion and sound tags (e.g., [sigh], <whisper>) plus few-shot voice cloning from a short reference clip, running efficiently on CPU. It is a fine-tune of Kyutai's Pocket TTS with an extended tokenizer and a distillation stage that bakes in classifier-free guidance, making it a practical, laptop-friendly TTS option with expressive control.

Target audience: ML engineers, hobbyists, and developers building on-device or offline speech interfaces

Repository: https://huggingface.co/BlueRoseLabs/Silencio-2

View on Hacker News