ChatTTS
A text-to-speech model built for dialogue scenarios such as an LLM assistant, generating multi-speaker conversational speech with control over prosody, laughter, pauses and interjections.
ChatTTS is optimized for dialogue-based tasks rather than single-utterance narration, producing natural, expressive speech and supporting multiple speakers for interactive conversations. It predicts and controls fine-grained prosodic features, including laughter, pauses and interjections, and the project reports better prosody than most open-source text-to-speech models it compares against.
The released open-source model was pretrained on 40,000 hours of Chinese and English audio without further fine-tuning; the main model used 100,000-plus hours. The model weights are released under a CC BY-NC 4.0 license restricted to educational and research use, while the code is published under AGPLv3+. To limit misuse, the released 40,000-hour model has added high-frequency noise and compressed audio quality, and the authors have trained an internal detection model they plan to open-source.
It is written in Python and supports English and Chinese, with more languages planned.