GitHub - 2noise/ChatTTS: ChatTTS is a generative speech model for daily dialogue.

What is ChatTTS?

ChatTTS is a generative speech model for daily dialogue.

How can I use ChatTTS?

You can use ChatTTS for text-to-speech synthesis, supporting both English and Chinese languages. Our model is trained with over 100,000 hours of data consisting of Chinese and English content.

Features of ChatTTS

ChatTTS is designed for dialogue-based tasks, allowing for natural and expressive speech synthesis. It accommodates multiple speakers, making interactive conversations easier. The model can anticipate and control detailed prosodic elements such as laughter, pauses, and interjections.

Roadmap

Our future plans involve releasing the 40,000-hour base model and spk_stats file as open source, sharing the VQ encoder and Lora training code, enabling streaming audio generation without refining the text, open-sourcing the 40,000-hour version with multi-emotion control, and potentially developing a ChatTTS.cpp.

FAQ

To generate a 30-second audio clip, a minimum of 4GB of GPU memory is needed. With a 4090 GPU, it can produce audio equivalent to about 7 semantic tokens per second. The Real-Time Factor (RTF) is approximately 0.3.