Demo: https://huggingface.co/spaces/Naozumi0512/Bert-VITS2-Cantonese
Note
This repository is no longer being actively maintained. The development of Bert-VITS2 in Cantonese has been moved to a new repository: hon9kon9ize/Bert-VITS2-Cantonese.
The new repository will continue improving the model, including training on longer duration audio data. Additionally, the model will maintain its ability to output speech in other languages, not limited to Cantonese.
We encourage users and contributors to follow the new repository for the latest updates, improvements, and contributions to the Cantonese TTS model.
To further improve the Cantonese TTS model, we are actively seeking contributions of audio data. If you have high-quality audio recordings in Cantonese (or other languages), please consider contributing them to the project.
Requirements for audio data contributions:
- Accurate transcriptions: Each audio file should have a corresponding text transcription that accurately represents the spoken content.
- Speaker separation: If the audio contains multiple speakers, please ensure that the speakers are separated and tagged appropriately.
By contributing your audio data, you'll be helping to enhance the model's performance and enable more natural and expressive speech synthesis. Meanwhile we will ensure that the final model cannot reproduce the original speaker's voice directly.
Original README content
VITS2 Backbone with multilingual bert
For quick guide, please refer to webui_preprocess.py.
简易教程请参见 webui_preprocess.py。
FishAudio下的全新自回归TTS Fish-Speech目前已经开始着手进行LLM大规模预训练工作。我们目前需要相对高质量的语音文件以进行最后的SFT（监督微调），因此在这里进行公开征集。 贡献的语音数据需要满足以下要求： 1.总时长>10h 2.需要是中文或者英文素材 3.不得存在伴奏，可以使用UVR素材 4.带文本优先，不带亦可 5.咬字清楚，无明显口音 如果您有符合上述条件的素材且愿意贡献于训练，请直接联系QQ 2225664821。 我们会对最终模型进行技术手段的处理，使之无法合成您素材对应的speaker音频。 您的贡献届时将会记录在贡献清单上，随release发布。 FishAudio Team
请注意，本项目核心思路来源于anyvoiceai/MassTTS 一个非常好的tts项目
MassTTS的演示demo为ai版峰哥锐评峰哥本人,并找回了在金三角失落的腰子
- anyvoiceai/MassTTS
- jaywalnut310/vits
- p0p4k/vits2_pytorch
- svc-develop-team/so-vits-svc
- PaddlePaddle/PaddleSpeech
- emotional-vits
- fish-speech
- Bert-VITS2-UI