Afula

@nickr61 Too many things need to be address right now, the only way to temporarily solve the problem is to git checkout a9eab236d7e8afa4d6205127904a385b2c43bb24 or previous version. But previous version have...

It's due to the dreambooth extension, it update GitPython to version 3.1.31 which breaks all update of other extension. Since update module requires GitPython to remain in version 3.1.27, remove...

Moreover, in order to make your extensions update as normal again, you may have to reinstall every extension one by one to make it updateable in the future, since gitpython...

> 不知道有没有人测试过，比如一句30s的音频，是做成1句30s，还是3句10s效果更好？也就是语音切分有没有什么讲究？ 在3s-10s之间选择，尽量每段音频的语速和情绪是一致的，例如一段悲伤的语音就不要包含转换成欢快情绪的片段了，最好不要有一句话说了一半就被截断的情况。超出10s没有意义，很多样本会直接被弃，根本不会生成频谱模型文件，也就不会被纳入到训练过程中。

Still having issues for check for update even rebuild venv and delete dreambooth completely, I guess someone can point out which package version is not compatible for the main version...

@Dasor92 Moreover, in order to make your extensions update as normal again, you may have to reinstall every extension one by one to make it updateable in the future, since...

Make sure you are eligible to access gpt-4, otherwise you'll need to change the modelName in makechain.ts to gpt-3.5-turbo.

> > Make sure you are eligible to access gpt-4, otherwise you'll need to change the modelName in makechain.ts to gpt-3.5-turbo. > > I guess yes, already upgrated my plan...

你参考一下[https://github.com/ywwwf/vits-mandarin-windows](url)这个项目吧，@coderyiyang，看一下代码改改，不过改的地方估计有点多。改train.py，首先要改nccl为gloo

> > 大佬我这边可能没描述清楚，单独的“劳模精神、劳动精神、工匠精神”没有问题，在长文本环境下就会出问题，比如：“大力弘扬劳模精神、劳动精神、工匠精神，让他倍感自豪”，主要想从切分或者替换的方法能不能解决，应该再训练毕竟麻烦。 > > [short.zip](https://github.com/RVC-Boss/GPT-SoVITS/files/15400849/short.zip) [long.zip](https://github.com/RVC-Boss/GPT-SoVITS/files/15400850/long.zip) > > 我试了一下的确是的，好像是会丢掉“工匠精神”，我改成“大力弘扬劳模精神。劳动精神。工匠精神。让他倍感自豪。”就没问题了，试试其他的切分方式，或者按标点符号切分。 @originwufish 因为精神两个字，在一句话里面出现了三遍，所以会吞。凡是重复的都会吞。比如叠词，语气词也会吞，比如：呵呵呵，第一个呵大概率会变为气声，当然如果你的语料中有大量的笑声作为样本是可以减缓这种现象。