#1414th of 15 in Voice
chattts
Text-to-speech model for dialogue, with laughter and pause control
- Stars
- 39.9k
- License
- AGPL-3.0
- Last commit
- Apr 2026
- Last release
- Apr 2026
- Language
- Python
38,454 on 24 Mar 202639,901 on 13 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
ChatTTS is a Python text-to-speech model trained for conversational speech in English and Chinese, with multi-speaker support and token-level control over laughter, pauses and interjections. It installs from PyPI, outputs 24 kHz audio, and ships a simple WebUI and command-line example. The released weights are a 40,000-hour pretrained model without SFT, licensed CC BY-NC 4.0 for research use.
Who it is for: Developers adding English or Chinese speech to LLM apps, for research use
Strengths
- Token-level prosody control: [laugh], [uv_break], [lbreak], plus sentence-level oral/laugh/break prompts
- Random speaker sampling with saved embeddings for timbre recovery
- Streaming generation and zero-shot inference code are included
- Runs with about 4 GB GPU memory for a 30-second clip; RTF around 0.3 on a 4090
Weaknesses
- Model weights are CC BY-NC 4.0, academic use only; no commercial use
- Only English and Chinese supported; English is described as experimental
- Autoregressive output is unstable: speaker drift and quality vary between samples
- Repo is the algorithm and examples; no packaged server or Docker image
What it needs
- GPU optional
- Needs PyTorch, torchaudio, vLLM (optional, Linux only)
- Models: ChatTTS 40k-hour base model (Hugging Face)
Also in Voice
See all 15| Rank | Project | Score |
|---|---|---|
| 1 | whisper.cppC/C++ port of OpenAI Whisper for local speech-to-text transcription | 74 out of 100 |
| 2 | VoiceboxLocal voice studio for cloning, TTS, dictation and agent speech | 69 out of 100 |
| 3 | Speech-to-SpeechModular voice-agent pipeline exposed through the OpenAI Realtime API | 66 out of 100 |
| #4 | Kokoro-FastAPIOpenAI-compatible text-to-speech server for the Kokoro-82M model | 65 out of 100 |
| #5 | Pocket TTSCPU-only text-to-speech model with voice cloning and streaming | 63 out of 100 |