[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-72061":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":8,"language":10,"languages":8,"totalLinesOfCode":8,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":16,"stars7d":16,"stars30d":17,"stars90d":15,"forks30d":15,"starsTrendScore":18,"compositeScore":19,"rankGlobal":8,"rankLanguage":8,"license":20,"archived":21,"fork":21,"defaultBranch":22,"hasWiki":21,"hasPages":21,"topics":23,"createdAt":8,"pushedAt":8,"updatedAt":25,"readmeContent":26,"aiSummary":27,"trendingCount":15,"starSnapshotCount":15,"syncStatus":17,"lastSyncTime":28,"discoverSource":29},72061,"MegaTTS3","bytedance\u002FMegaTTS3","bytedance",null,"","Python",6084,471,50,86,0,1,2,3,39.02,"Apache License 2.0",false,"main",[24],"research","2026-06-12 02:02:58","\u003Cdiv align=\"center\">\n    \u003Ch1>\n    MegaTTS 3 \u003Cimg src=\".\u002Fassets\u002Ffig\u002FHi.gif\" width=\"40px\">\n    \u003C\u002Fh1>\n    \u003Cp>\n    Official PyTorch Implementation\u003Cbr>\n    \u003C\u002Fp>\n\u003C\u002Fdiv>\n\u003Cdiv align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fspaces\u002FByteDance\u002FMegaTTS3\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHugging%20Face-Space%20Demo-yellow\" alt=\"Hugging Face\">\u003C\u002Fa>\n    \u003Ca href=\"#\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPlatform-linux-lightgrey\" alt=\"version\">\u003C\u002Fa>\n    \u003Ca href=\"#\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPython-3.10-brightgreen\" alt=\"version\">\u003C\u002Fa>\n    \u003Ca href=\"#\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPyTorch-2.6.0-orange\" alt=\"python\">\u003C\u002Fa>\n    \u003Ca href=\"#\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-Apache%202.0-blue.svg\" alt=\"mit\">\u003C\u002Fa>\n\u003C\u002Fdiv>\n\u003Cdiv align=\"center\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FBytedance-%230077B5.svg?&style=flat-square&logo=bytedance&logoColor=white\" \u002F>\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FZhejiang University-%230077B5.svg?&style=flat-square&logo=data:image\u002Fsvg+xml;base64,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&logoColor=white\" \u002F>\n\u003C\u002Fdiv>\n\n## Key features\n- 🚀**Lightweight and Efficient:** The backbone of the TTS Diffusion Transformer has only 0.45B parameters.\n- 🎧**Ultra High-Quality Voice Cloning:** You can try our model at [Huggingface Demo](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002FByteDance\u002FMegaTTS3)🎉. The .wav and .npy files can be found at [link1](https:\u002F\u002Fdrive.google.com\u002Fdrive\u002Ffolders\u002F1QhcHWcy20JfqWjgqZX1YM3I6i9u4oNlr?usp=sharing). Submit a sample (.wav format, \u003C 24s, and please do not contain space in filename) on [link2](https:\u002F\u002Fdrive.google.com\u002Fdrive\u002Ffolders\u002F1gCWL1y_2xu9nIFhUX_OW5MbcFuB7J5Cl?usp=sharing) to receive .npy voice latents you can use locally.\n- 🌍**Bilingual Support:** Supports both Chinese and English, and code-switching.\n- ✍️**Controllable:** Supports accent intensity control ✅ and fine-grained pronunciation\u002Fduration adjustment (coming soon).\n\n[MegaTTS 3 Demo Video](https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F0174c111-f392-4376-a34b-0b5b8164aacc)\n\n\u003Cdiv style='width:100%;text-align:center'>\n\u003Cimg src=\".\u002Fassets\u002Ffig\u002Ftable_tts.png\" width=\"550px\">\n\u003C\u002Fdiv>\n\n## 🎯Roadmap\n\n- **[2025-03-22]** Our project has been released!\n\n\n## Installation\n``` sh\n# Clone the repository\ngit clone https:\u002F\u002Fgithub.com\u002Fbytedance\u002FMegaTTS3\ncd MegaTTS3\n```\n**Requirements (for Linux)**\n``` sh\n\n# Create a python 3.10 conda env (you could also use virtualenv)\nconda create -n megatts3-env python=3.10\nconda activate megatts3-env\npip install -r requirements.txt\n\n# Set the root directory\nexport PYTHONPATH=\"\u002Fpath\u002Fto\u002FMegaTTS3:$PYTHONPATH\"\n\n# [Optional] Set GPU\nexport CUDA_VISIBLE_DEVICES=0\n\n# If you encounter bugs with pydantic in inference, you should check if the versions of pydantic and gradio are matched.\n# [Note] if you encounter bugs related with httpx, please check that whether your environmental variable \"no_proxy\" has patterns like \"::\"\n```\n\n**Requirements (for Windows)**\n``` sh\n# [The Windows version is currently under testing]\n# Comment below dependence in requirements.txt:\n# # WeTextProcessing==1.0.4.1\n\n# Create a python 3.10 conda env (you could also use virtualenv)\nconda create -n megatts3-env python=3.10\nconda activate megatts3-env\npip install -r requirements.txt\nconda install -y -c conda-forge pynini==2.1.5\npip install WeTextProcessing==1.0.3\n\n# [Optional] If you want GPU inference, you may need to install specific version of PyTorch for your GPU from https:\u002F\u002Fpytorch.org\u002F.\npip install torch torchvision torchaudio --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu126\n\n# [Note] if you encounter bugs related with `ffprobe` or `ffmpeg`, you can install it through `conda install -c conda-forge ffmpeg`\n\n# Set environment variable for root directory\nset PYTHONPATH=\"C:\\path\\to\\MegaTTS3;%PYTHONPATH%\" # Windows\n$env:PYTHONPATH=\"C:\\path\\to\\MegaTTS3;%PYTHONPATH%\" # Powershell on Windows\nconda env config vars set PYTHONPATH=\"C:\\path\\to\\MegaTTS3;%PYTHONPATH%\" # For conda users\n\n# [Optional] Set GPU\nset CUDA_VISIBLE_DEVICES=0 # Windows\n$env:CUDA_VISIBLE_DEVICES=0 # Powershell on Windows\n\n```\n\n**Requirements (for Docker)**\n``` sh\n# [The Docker version is currently under testing]\n# ! You should download the pretrained checkpoint before running the following command\ndocker build . -t megatts3:latest\n\n# For GPU inference\ndocker run -it -p 127.0.0.1:7929:7929 --gpus all -e CUDA_VISIBLE_DEVICES=0 megatts3:latest\n# For CPU inference\ndocker run -it -p 127.0.0.1:7929:7929  megatts3:latest\n\n# Visit http:\u002F\u002F127.0.0.1:7860\u002F for gradio.\n```\n\n\n**Model Download**\n\nThe pretrained checkpoint can be found at [Google Drive](https:\u002F\u002Fdrive.google.com\u002Fdrive\u002Ffolders\u002F1CidiSqtHgJTBDAHQ746_on_YR0boHDYB?usp=sharing) or [Huggingface](https:\u002F\u002Fhuggingface.co\u002FByteDance\u002FMegaTTS3). Please download them and put them to ``.\u002Fcheckpoints\u002Fxxx``.\n\n> [!IMPORTANT]  \n> 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](https:\u002F\u002Fdrive.google.com\u002Fdrive\u002Ffolders\u002F1QhcHWcy20JfqWjgqZX1YM3I6i9u4oNlr?usp=sharing) 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.\n> \n> This project is primarily intended for academic purposes. For academic datasets requiring evaluation, you may upload them to the voice request queue in [link2](https:\u002F\u002Fdrive.google.com\u002Fdrive\u002Ffolders\u002F1gCWL1y_2xu9nIFhUX_OW5MbcFuB7J5Cl?usp=sharing) (within 24s for each clip). After verifying that your uploaded voices are free from safety issues, we will upload their latent files to [link1](https:\u002F\u002Fdrive.google.com\u002Fdrive\u002Ffolders\u002F1QhcHWcy20JfqWjgqZX1YM3I6i9u4oNlr?usp=sharing) as soon as possible.\n> \n> In the coming days, we will also prepare and release the latent representations for some common TTS benchmarks.\n\n## Inference\n\n**Command-Line Usage (Standard)**\n``` bash\n# p_w (intelligibility weight), t_w (similarity weight). Typically, prompt with more noises requires higher p_w and t_w\npython tts\u002Finfer_cli.py --input_wav 'assets\u002FChinese_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 .\u002Fgen\n\n# As long as audio volume and pronunciation are appropriate, increasing --t_w within reasonable ranges (2.0~5.0)\n# will increase the generated speech's expressiveness and similarity (especially for some emotional cases).\npython tts\u002Finfer_cli.py --input_wav 'assets\u002FEnglish_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 .\u002Fgen --p_w 2.0 --t_w 3.0\n```\n**Command-Line Usage (for TTS with Accents)**\n``` bash\n# 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. \n# t_w (similarity weight) is typically set 0–3 points higher than p_w for optimal results.\n# Useful for accented TTS or solving the accent problems in cross-lingual TTS.\npython tts\u002Finfer_cli.py --input_wav 'assets\u002FEnglish_prompt.wav' --input_text '这是一条有口音的音频。' --output_dir .\u002Fgen --p_w 1.0 --t_w 3.0\n\npython tts\u002Finfer_cli.py --input_wav 'assets\u002FEnglish_prompt.wav' --input_text '这条音频的发音标准一些了吗？' --output_dir .\u002Fgen --p_w 2.5 --t_w 2.5\n```\n\n**Web UI Usage**\n``` bash\n# We also support cpu inference, but it may take about 30 seconds (for 10 inference steps).\npython tts\u002Fgradio_api.py\n```\n\n## Submodules\n> [!TIP]\n> In addition to TTS, some submodules in this project may also have additional usages.\n> See ``.\u002Ftts\u002Ffrontend_fuction.py`` and ``.\u002Ftts\u002Finfer_cli.py`` for example code.\n\n### Aligner\n**Description:** a robust speech-text aligner model trained using pseudo-labels generated by a large number of MFA expert models.\n\n**Usage**: 1) Prepare the finetuning dataset for our model; 2) Filter the large-scale speech dataset (if the aligner fails to align a certain speech clip, it is likely to be noisy); 3) Phoneme recognition; 4) Speech segmentation.\n\n### Graphme-to-Phoneme Model\n**Description:** a Qwen2.5-0.5B model finetuned for robust graphme-to-phoneme conversion.\n\n**Usage**: Graphme-to-phoneme conversion.\n\n### WaveVAE\n**Description:** a strong waveform VAE that can compress 24 kHz speeche into 25 Hz acoustic latent and reconstruct the original wave almost losslessly.\n\n**Usage:** 1) Acoustic latents can provide a more compact and discriminative training target for speech synthesis models compared to mel-spectrograms, accelerating convergence; 2) Used as acoustic latents for voice conversion; 3) High-quality vocoder.\n\n\u003Cdiv style='width:100%;text-align:center'>\n\u003Cimg src=\".\u002Fassets\u002Ffig\u002Ftable_wavvae.png\" width=\"650px\">\n\u003C\u002Fdiv>\n\n\n## Security\nIf you discover a potential security issue in this project, or think you may\nhave discovered a security issue, we ask that you notify Bytedance Security via our [security center](https:\u002F\u002Fsecurity.bytedance.com\u002Fsrc) or [sec@bytedance.com](sec@bytedance.com).\n\nPlease do **not** create a public GitHub issue.\n\n## License\nThis project is licensed under the [Apache-2.0 License](LICENSE).\n\n## Citation\nThis repo contains forced-align version of `Sparse 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.\n```\n@article{jiang2025sparse,\n  title={Sparse Alignment Enhanced Latent Diffusion Transformer for Zero-Shot Speech Synthesis},\n  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},\n  journal={arXiv preprint arXiv:2502.18924},\n  year={2025}\n}\n\n@article{ji2024wavtokenizer,\n  title={Wavtokenizer: an efficient acoustic discrete codec tokenizer for audio language modeling},\n  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},\n  journal={arXiv preprint arXiv:2408.16532},\n  year={2024}\n}\n```\n","MegaTTS 3 是一个基于 PyTorch 的文本转语音（TTS）系统。其核心功能包括使用轻量级且高效的扩散变换器作为骨干网络，参数量仅为0.45亿，能够生成高质量的语音克隆，并支持中英文双语及代码切换。该项目还提供了对口音强度的控制以及未来将实现的发音和时长微调功能。适用于需要高质量语音合成的研究场景或应用开发，如虚拟助手、有声读物制作等。","2026-06-11 03:40:10","high_star"]