ChatTTS: audio generator — what it does and what you need to run it

ChatTTS is an open source audio generator from the 2noise team. It tackles the text-to-audio task: converting text into speech oriented toward everyday conversation. The authors describe the model as a generative speech model for daily communication.

What it does

As described by the author, ChatTTS generates speech from text. The model supports Chinese and English — this is indicated in the repository tags. The project description mentions working with semantic tokens: a 30-second audio clip requires at least 4 GB of VRAM, and on a 4090 GPU generation runs at roughly 7 semantic tokens per second. The real-time factor (RTF) is about 0.3 — this is the developer's claim, not an independent measurement.

The repository is tagged with topics such as agent, chat, llm, text-to-speech, and others, pointing to the model's focus on dialogue systems and agents. The latest release is v0.2.5.

What you need to run it

The code is written in Python and distributed under the AGPL-3.0 license. The model weights are available under the cc-by-nc-4.0 license — this means non-commercial use. The largest weight file is 0.8 GB, and all files together total 1.1 GB.

A GPU is required. As described by the author, the minimum VRAM is 4 GB for a 30-second clip. The integration library is called chat_tts. The code and weights are available on GitHub and Hugging Face.

Who it suits

ChatTTS may be of interest to those building dialogue systems, chatbots, or voice agents in Chinese or English. The open source code allows studying the implementation and fine-tuning the model for your own tasks, provided the AGPL-3.0 license works for your project.

The model is not suitable for commercial projects without separate approval: the weights are distributed under the non-commercial cc-by-nc-4.0 license. It is also not designed for systems without a discrete GPU — the authors state the VRAM requirement but do not describe CPU-only operation.

ChatTTS is an open tool for dialogue speech synthesis with a focus on Chinese and English. It requires a GPU and suits non-commercial projects that need real-time or near-real-time speech generation.