Views
No views yet
1# Clone the repository
2git clone https://github.com/bytedance/MegaTTS3
3cd MegaTTS3huggingface-cli download ByteDance/MegaTTS3 --local-dir ./checkpoints --local-dir-use-symlinks False1# Create a python 3.10 conda env (you could also use virtualenv)
2conda create -n megatts3-env python=3.10
3conda activate megatts3-env
4pip install -r requirements.txt
5
6# Set the root directory
7export PYTHONPATH="/path/to/MegaTTS3:$PYTHONPATH"
8
9# [Optional] Set GPU
10export CUDA_VISIBLE_DEVICES=0
11
12# If you encounter bugs with pydantic in inference, you should check if the versions of pydantic and gradio are matched.
13# [Note] if you encounter bugs related with httpx, please check that whether your environmental variable "no_proxy" has patterns like "::"1# [The Windows version is currently under testing]
2# Comment below dependence in requirements.txt:
3# # WeTextProcessing==1.0.4.1
4
5# Create a python 3.10 conda env (you could also use virtualenv)
6conda create -n megatts3-env python=3.10
7conda activate megatts3-env
8pip install -r requirements.txt
9conda install -y -c conda-forge pynini==2.1.5
10pip install WeTextProcessing==1.0.3
11
12# [Optional] If you want GPU inference, you may need to install specific version of PyTorch for your GPU from https://pytorch.org/.
13pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126
14
15# [Note] if you encounter bugs related with `ffprobe` or `ffmpeg`, you can install it through `conda install -c conda-forge ffmpeg`
16
17# Set environment variable for root directory
18set PYTHONPATH="C:\path\to\MegaTTS3;%PYTHONPATH%" # Windows
19$env:PYTHONPATH="C:\path\to\MegaTTS3;%PYTHONPATH%" # Powershell on Windows
20conda env config vars set PYTHONPATH="C:\path\to\MegaTTS3;%PYTHONPATH%" # For conda users
21
22# [Optional] Set GPU
23set CUDA_VISIBLE_DEVICES=0 # Windows
24$env:CUDA_VISIBLE_DEVICES=0 # Powershell on Windows1# [The Docker version is currently under testing]
2# ! You should download the pretrained checkpoint before running the following command
3docker build . -t megatts3:latest
4
5# For GPU inference
6docker run -it -p 7929:7929 --gpus all -e CUDA_VISIBLE_DEVICES=0 megatts3:latest
7# For CPU inference
8docker run -it -p 7929:7929 megatts3:latest
9
10# Visit http://0.0.0.0:7929/ for gradio.[!TIP] [IMPORTANT] For security issues, we do not upload the parameters of WaveVAE encoder to the above links. You can only use the pre-extracted latents from link1 for inference. If you want to synthesize speech for speaker A, you need "A.wav" and "A.npy" in the same directory. If you have any questions or suggestions for our model, please email us.This project is primarily intended for academic purposes. For academic datasets requiring evaluation, you may upload them to the voice request queue in link2 (within 24s for each clip). After verifying that your uploaded voices are free from safety issues, we will upload their latent files to link1 as soon as possible.In the coming days, we will also prepare and release the latent representations for some common TTS benchmarks.
1# p_w (intelligibility weight), t_w (similarity weight). Typically, prompt with more noises requires higher p_w and t_w
2python tts/infer_cli.py --input_wav 'assets/Chinese_prompt.wav' --input_text "另一边的桌上,一位读书人嗤之以鼻道,'佛子三藏,神子燕小鱼是什么样的人物,李家的那个李子夜如何与他们相提并论？'" --output_dir ./gen
3
4# As long as audio volume and pronunciation are appropriate, increasing --t_w within reasonable ranges (2.0~5.0)
5# will increase the generated speech's expressiveness and similarity (especially for some emotional cases).
6python tts/infer_cli.py --input_wav 'assets/English_prompt.wav' --input_text 'As his long promised tariff threat turned into reality this week, top human advisers began fielding a wave of calls from business leaders, particularly in the automotive sector, along with lawmakers who were sounding the alarm.' --output_dir ./gen --p_w 2.0 --t_w 3.01# When p_w (intelligibility weight) ≈ 1.0, the generated audio closely retains the speaker’s original accent. As p_w increases, it shifts toward standard pronunciation.
2# t_w (similarity weight) is typically set 0–3 points higher than p_w for optimal results.
3# Useful for accented TTS or solving the accent problems in cross-lingual TTS.
4python tts/infer_cli.py --input_wav 'assets/English_prompt.wav' --input_text '这是一条有口音的音频。' --output_dir ./gen --p_w 1.0 --t_w 3.0
5
6python tts/infer_cli.py --input_wav 'assets/English_prompt.wav' --input_text '这条音频的发音标准一些了吗？' --output_dir ./gen --p_w 2.5 --t_w 2.51# We also support cpu inference, but it may take about 30 seconds (for 10 inference steps).
2python tts/gradio_api.pySparse Alignment Enhanced Latent Diffusion Transformer for Zero-Shot Speech Synthesis and the WavVAE is mainly based on Wavtokenizer: an efficient acoustic discrete codec tokenizer for audio language modeling. Compared to the model described in paper, the repository includes additional models. These models not only enhance the stability and cloning capabilities of the algorithm but can also be independently utilized to serve a wider range of scenarios.@article{jiang2025sparse,
title={Sparse Alignment Enhanced Latent Diffusion Transformer for Zero-Shot Speech Synthesis},
author={Jiang, Ziyue and Ren, Yi and Li, Ruiqi and Ji, Shengpeng and Ye, Zhenhui and Zhang, Chen and Jionghao, Bai and Yang, Xiaoda and Zuo, Jialong and Zhang, Yu and others},
journal={arXiv preprint arXiv:2502.18924},
year={2025}
}
@article{ji2024wavtokenizer,
title={Wavtokenizer: an efficient acoustic discrete codec tokenizer for audio language modeling},
author={Ji, Shengpeng and Jiang, Ziyue and Wang, Wen and Chen, Yifu and Fang, Minghui and Zuo, Jialong and Yang, Qian and Cheng, Xize and Wang, Zehan and Li, Ruiqi and others},
journal={arXiv preprint arXiv:2408.16532},
year={2024}
}