XXXXRT666
XXXXRT666
工作路径问题？需要在GPT_SoVITS上层目录启动
> 虽然似乎合理了，但是 #950 ? 先改成repeat占比0.9了
> 虽然似乎合理了，但是 #950 ? 他说的更像是取[0]的原因，还需要进一步实验
句子太长或者是空的都会导致，参考太长也会
没看到报错，如果明确是warning导致的问题可以环境变量关闭这个warning
> did you install the miniconda in x86_64 way? check your `conda-info` info > > if x86_64, it may cause problem in CTtranslated2 to stop the faster whisper. I think...
https://github.com/SYSTRAN/faster-whisper/issues/345 this might help
Well, G2P is all you need. Find a suitable phonetic symbol, and use G2P to convert your label to phones.
Maybe you can train it without pre-trained models, or just, train a pre-trained model with your large dataset. Have a look at this #67