Verified project record

2noise/ChatTTS

ChatTTS is a text-to-speech model designed for dialogue scenarios, providing fine-grained prosodic control for features like laughter and pauses. It supports English and Chinese, operates under a non-commercial academic license, and can be run locally via a WebUI, command line, or as a library.

Project overview

It specifically targets dialogue-based tasks with token-level prosodic controls such as laughter, pauses, and interjections.

Project type: Audio & Speech
Use cases: Audio & Speech
Deployment: Python environment
License: AGPL-3.0

Best for

Researchers and developers building academic or educational dialogue-based speech synthesis applications.

Key capabilities

Optimized for dialogue-based tasks, enabling natural and expressive speech synthesis with multiple speakers support.
Predicts and controls fine-grained prosodic features including laughter, pauses, and interjections via special tokens.
Capability to control additional emotions during speech synthesis.
Launches a local WebUI for interaction with the model.
Executes inference from the command line and saves audio files.

Limitations and risks

The released model is intended for educational and research use.
Model stability is not good enough, with documented issues such as multiple speakers or poor audio quality.
Only English and Chinese are currently supported.
Only laugh, uv_break, and lbreak token-level control units are available.
The model is published under CC BY-NC 4.0 license and should not be used for commercial purposes.
A small amount of high-frequency noise was added during training, and audio quality was compressed using MP3 to prevent malicious use.

Getting started

Setup requires a Python environment and model weights download. Users can install the package via pip or clone the repository, then launch the local WebUI or execute inference via CLI script.

Alternatives and comparisons

A deep learning toolkit for Text-to-Speech generation that supports training, fine-tuning, and synthesizing speech in over 1100 languages.
A tokenizer-free text-to-speech model supporting 30 languages, 48kHz audio output, voice design, and voice cloning.
A multilingual text-to-speech system that generates speech in over 80 languages and clones voices using short reference samples.

Project comparisons

Evidence and sources

Open original GitHub project