ChatTTS
Text-to-speech model built for dialogue, trained on English and Chinese audio, with control over laughter, pauses and interjections.
- GitHub stars
- 40k
- Last commit
- 7 mo ago
- Latest release
- v0.2.5
- Licence
- AGPL-3.0
- Self-hosted
- Yes

ChatTTS is a generative speech model aimed at dialogue scenarios such as LLM assistants. The repository holds the algorithm infrastructure and some simple examples; end-user products built on the model are collected in a community-maintained index repo. English and Chinese are the supported languages, with more marked as coming.
The main model was trained on more than 100,000 hours of Chinese and English audio, but the version published on HuggingFace is a 40,000-hour pre-trained model without SFT. The roadmap lists the 40k-hour base model, streaming audio generation and the DVAE encoder with zero-shot inference code as released, while multi-emotion control and a ChatTTS.cpp port remain open.
It is a Python project: install from PyPI with pip, from GitHub, or in a conda environment on Python 3.11, then start the web UI with examples/web/webui.py or generate MP3 files from the command line. The code is under AGPLv3+, while the model is under CC BY-NC 4.0 for educational and research use. The authors added high-frequency noise and MP3 compression to the 40,000-hour model to deter misuse.
Key features
- Speech synthesis tuned for dialogue
- Multiple speakers
- Controls laughter, pauses and interjections
- Web UI and command-line inference
- Installable from PyPI
- Trained on Chinese and English audio
- Pretrained models provided

















