#2835·GPT-SoVITS

[License/Provenance] Clarification for G2PWModel.zip and complete offline dependencies

Author: jasonll0205ll-rgbCreated Sep 2, 2026Updated Sep 2, 2026

您好,感谢维护 GPT-SoVITS。为了在可商业发布、完全离线的中文 TTS 流程中合规使用官方文档推荐的 G2PW 模型,希望确认以下公开制品的许可与来源信息。

审计对象:

该 ZIP 内没有 LICENSE、NOTICE 或来源清单。我们看到 Hugging Face 仓库卡标注 MIT,GPT-SoVITS 代码仓库为 MIT,上游 g2pW 代码为 Apache-2.0;但这些仓库级/代码级声明不足以确定 ZIP 内模型权重和数据衍生文件的逐制品许可。

烦请维护者确认:

  1. 上述固定 revision 的 G2PWModel.zip 内全部 8 个文件(尤其是 g2pW.onnx、字符/拼音词典及 MONOPHONIC_CHARS.txtPOLYPHONIC_CHARS.txt)分别适用什么许可证?仓库卡的 MIT 是否明确覆盖这些文件?
  2. 维护者/发布者是否有权按该许可证发布这些权重、词典和训练衍生数据?它们的原始来源、训练模型/checkpoint 与数据集许可是什么?
  3. 是否允许商业使用、修改和再分发该 ZIP 或其解压文件?如允许,需要保留哪些版权、署名或 NOTICE?
  4. config.py 指向 bert-base-chinese。完全离线运行时,官方推荐的完整 model_source 制品、固定版本/哈希、许可证及目录放置方式是什么?

如已有权威说明,请直接提供对应链接即可。得到明确答复前,我们会保持 fail-closed,不安装或运行该制品。


Hello, and thank you for maintaining GPT-SoVITS. We are auditing the documented G2PW dependency for a commercially publishable, fully offline Chinese TTS workflow.

Could you please clarify, for the exact artifact and revision above:

  1. The license covering each archived file, especially g2pW.onnx, dictionaries, and mono/polyphonic character lists, and whether the repository-level MIT metadata explicitly covers them.
  2. The provenance and redistribution authority for the weights and derived data, including the source checkpoint/model and dataset licenses.
  3. Whether commercial use, modification, and redistribution are permitted, and any required attribution/NOTICE.
  4. The official pinned, licensed, complete offline model_source package for bert-base-chinese, including version/hash and placement instructions.

Until this is clarified, we will keep the artifact blocked rather than infer artifact-level rights from repository/code-level licenses.