ChatTTS README

The project's README from github.com/2noise/ChatTTS. As one markdown file: README.md.

2noise%2FChatTTS | Trendshift

ChatTTS

A generative speech model for daily dialogue.

Licence PyPI

Huggingface Open In Colab Discord

English | 简体中文 | 日本語 | Русский | Español | Français | 한국어

Introduction

[!Note] This repo contains the algorithm infrastructure and some simple examples.

[!Tip] For the extended end-user products, please refer to the index repo Awesome-ChatTTS maintained by the community.
You can find a diagram visualization of the codebase here.

ChatTTS is a text-to-speech model designed specifically for dialogue scenarios such as LLM assistant.

Supported Languages

Highlights

You can refer to this video on Bilibili for the detailed description.

  1. Conversational TTS: ChatTTS is optimized for dialogue-based tasks, enabling natural and expressive speech synthesis. It supports multiple speakers, facilitating interactive conversations.
  2. Fine-grained Control: The model could predict and control fine-grained prosodic features, including laughter, pauses, and interjections.
  3. Better Prosody: ChatTTS surpasses most of open-source TTS models in terms of prosody. We provide pretrained models to support further research and development.

Dataset & Model

[!Important] The released model is for academic purposes only.

Roadmap

Licenses

The Code

The code is published under AGPLv3+ license.

The model

The model is published under CC BY-NC 4.0 license. It is intended for educational and research use, and should not be used for any commercial or illegal purposes. The authors do not guarantee the accuracy, completeness, or reliability of the information. The information and data used in this repo, are for academic and research purposes only. The data obtained from publicly available sources, and the authors do not claim any ownership or copyright over the data.

Disclaimer

ChatTTS is a powerful text-to-speech system. However, it is very important to utilize this technology responsibly and ethically. To limit the use of ChatTTS, we added a small amount of high-frequency noise during the training of the 40,000-hour model, and compressed the audio quality as much as possible using MP3 format, to prevent malicious actors from potentially using it for criminal purposes. At the same time, we have internally trained a detection model and plan to open-source it in the future.

Contact

GitHub issues/PRs are always welcomed.

Formal Inquiries

For formal inquiries about the model and roadmap, please contact us at [email protected].

Online Chat

1. QQ Group (Chinese Social APP)
2. Discord Server

Join by clicking here.

Get Started

Clone Repo

git clone https://github.com/2noise/ChatTTS
cd ChatTTS

Install requirements

1. Install Directly

pip install --upgrade -r requirements.txt

2. Install from conda

conda create -n chattts python=3.11
conda activate chattts
pip install -r requirements.txt

Optional: Install vLLM (Linux only)

pip install safetensors vllm==0.2.7 torchaudio

Unrecommended Optional: Install TransformerEngine if using NVIDIA GPU (Linux only)

[!Warning] DO NOT INSTALL! The adaptation of TransformerEngine is currently under development and CANNOT run properly now. Only install it on developing purpose. See more details on at #672 #676

[!Note] The installation process is very slow.

pip install git+https://github.com/NVIDIA/TransformerEngine.git@stable

Unrecommended Optional: Install FlashAttention-2 (mainly NVIDIA GPU)

[!Warning] DO NOT INSTALL! Currently the FlashAttention-2 will slow down the generating speed according to this issue. Only install it on developing purpose.

[!Note] See supported devices at the Hugging Face Doc.

pip install flash-attn --no-build-isolation

Quick Start

Make sure you are under the project root directory when you execute these commands below.

1. Launch WebUI

python examples/web/webui.py

2. Infer by Command Line

It will save audio to ./output_audio_n.mp3

python examples/cmd/run.py "Your text 1." "Your text 2."

Installation

  1. Install the stable version from PyPI
pip install ChatTTS
  1. Install the latest version from GitHub
pip install git+https://github.com/2noise/ChatTTS
  1. Install from local directory in dev mode
pip install -e .

Basic Usage

import ChatTTS
import torch
import torchaudio

chat = ChatTTS.Chat()
chat.load(compile=False) # Set to True for better performance

texts = ["PUT YOUR 1st TEXT HERE", "PUT YOUR 2nd TEXT HERE"]

wavs = chat.infer(texts)

for i in range(len(wavs)):
    """
    In some versions of torchaudio, the first line works but in other versions, so does the second line.
    """
    try:
        torchaudio.save(f"basic_output{i}.wav", torch.from_numpy(wavs[i]).unsqueeze(0), 24000)
    except:
        torchaudio.save(f"basic_output{i}.wav", torch.from_numpy(wavs[i]), 24000)

Advanced Usage

###################################
# Sample a speaker from Gaussian.

rand_spk = chat.sample_random_speaker()
print(rand_spk) # save it for later timbre recovery

params_infer_code = ChatTTS.Chat.InferCodeParams(
    spk_emb = rand_spk, # add sampled speaker 
    temperature = .3,   # using custom temperature
    top_P = 0.7,        # top P decode
    top_K = 20,         # top K decode
)

###################################
# For sentence level manual control.

# use oral_(0-9), laugh_(0-2), break_(0-7) 
# to generate special token in text to synthesize.
params_refine_text = ChatTTS.Chat.RefineTextParams(
    prompt='[oral_2][laugh_0][break_6]',
)

wavs = chat.infer(
    texts,
    params_refine_text=params_refine_text,
    params_infer_code=params_infer_code,
)

###################################
# For word level manual control.

text = 'What is [uv_break]your favorite english food?[laugh][lbreak]'
wavs = chat.infer(text, skip_refine_text=True, params_refine_text=params_refine_text,  params_infer_code=params_infer_code)
"""
In some versions of torchaudio, the first line works but in other versions, so does the second line.
"""
try:
    torchaudio.save("word_level_output.wav", torch.from_numpy(wavs[0]).unsqueeze(0), 24000)
except:
    torchaudio.save("word_level_output.wav", torch.from_numpy(wavs[0]), 24000)

Example: self introduction

 
inputs_en = """
chat T T S is a text to speech model designed for dialogue applications. 
[uv_break]it supports mixed language input [uv_break]and offers multi speaker 
capabilities with precise control over prosodic elements like 
[uv_break]laughter[uv_break][laugh], [uv_break]pauses, [uv_break]and intonation. 
[uv_break]it delivers natural and expressive speech,[uv_break]so please
[uv_break] use the project responsibly at your own risk.[uv_break]
""".replace('\n', '') # English is still experimental.

params_refine_text = ChatTTS.Chat.RefineTextParams(
    prompt='[oral_2][laugh_0][break_4]',
)

audio_array_en = chat.infer(inputs_en, params_refine_text=params_refine_text)
torchaudio.save("self_introduction_output.wav", torch.from_numpy(audio_array_en[0]), 24000)

male speaker

female speaker

male speaker

female speaker

FAQ

1. How much VRAM do I need? How about infer speed?

For a 30-second audio clip, at least 4GB of GPU memory is required. For the 4090 GPU, it can generate audio corresponding to approximately 7 semantic tokens per second. The Real-Time Factor (RTF) is around 0.3.

2. Model stability is not good enough, with issues such as multi speakers or poor audio quality.

This is a problem that typically occurs with autoregressive models (for bark and valle). It's generally difficult to avoid. One can try multiple samples to find a suitable result.

3. Besides laughter, can we control anything else? Can we control other emotions?

In the current released model, the only token-level control units are [laugh], [uv_break], and [lbreak]. In future versions, we may open-source models with additional emotional control capabilities.

Acknowledgements

Special Appreciation

Thanks to all contributors for their efforts

contributors

counter

ChatTTS (2noise/ChatTTS)

Searching Database…

39.9K ·GitHub
ChatTTS
What is this?ChatTTS is an open-source text-to-speech model built for conversational speech — the natural-sounding voice layer for things like LLM assistants — speaking English and Chinese.
So is this a finished app, or something I build with?Developer material first: a pip-installable Python package around pretrained weights, plus a bundled WebUI and CLI for quick generation. The repo itself is the algorithm infrastructure with simple examples; end-user products live in the community Awesome-ChatTTS index.
How do I get it generating speech?Install it with pip (stable from PyPI, latest from GitHub, or local dev mode; conda is also documented), then from the project root launch the WebUI or run command-line inference — output is saved as an MP3.
Is there a catch with the audio quality?Yes — on purpose: to deter criminal misuse, the authors added high-frequency noise during training and compressed the output to MP3 quality. A detection model is planned but not yet released.
Can I use this in a commercial product?No. The code is AGPLv3+, and the model weights are CC BY-NC 4.0 — educational and research use only, commercial use is explicitly off the table.
What can I control in the generated speech?Fine-grained prosody — laughter, pauses and interjections. In the released checkpoint the token-level control units are [laugh], [uvbreak] and [lbreak]; broader multi-emotion control is still on the roadmap.
Why do some generations sound off or switch speakers?That is typical autoregressive-model instability, the same family as 'bark' and 'valle', and hard to avoid entirely — the README's advice is to generate several samples and keep a good one.
Are there finished apps built on it?Not in this repo — it ships the model and simple examples. Ready-made end-user products are collected in the community-maintained Awesome-ChatTTS index, and there are QQ groups and a Discord server.
What is already open-sourced and what is coming?Done: the 40k-hour base model with spkstats, streaming audio generation, and the DVAE encoder with zero-shot inferring code. Pending: multi-emotion control and a ChatTTS.cpp C++ port.
What are similar projects?csm, CosyVoice, dia, AudioGPT, bark, Speech, fish-speech, SoulX-Podcast
Which topics is it in?torch, llm-agent, conversational-ai, voice-assistant, audio-generation, english, voice, chinese-language, webui, agplv3, tts, chat
Creator
2noise
Size
10.5 MB
Stars
39.9K
Updated
CommitApr 10Releasev0.2.5Apr 10
Age
2y
Commits · 90 Days