KittenML releases Kitten-TTS 2, an open-source 1.58-bit TTS model with real-time CPU inference

·2026.10.22 09:33

Key point

The 1.7B parameter model supports zero-shot voice cloning in 20+ languages and expressive emphasis control without requiring a GPU.

KittenML released Kitten-TTS 2, the world's first 1.58-bit TTS model, with code and weights available open-source on GitHub and Hugging Face.
The model uses ternary weights (-1, 0, +1) with a 1.7B parameter count, enabling real-time inference on CPU and consumer GPUs without specialized hardware.
Features include zero-shot voice cloning from less than 5 seconds of reference audio, support for 20+ languages, and control over expressive delivery and word emphasis.
The release represents a complete architectural change from version 0.8, offering higher quality and natural-sounding speech as an open-source alternative to commercial services like ElevenLabs.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.