ConsistencyVC / ConsistencyVC-voive-conversion

Using joint training speaker encoder with consistency loss to achieve cross-lingual voice conversion and expressive voice conversion
https://consistencyvc.github.io/ConsistencyVC-demo-page
MIT License
134 stars 22 forks source link

必须要运行10K个epoch吗 #12

Closed tppqt closed 1 year ago

tppqt commented 1 year ago

用300条语音数据,微调了2200个epoch,发现转换后全是电流声,是语音数据太少了还是说是训练epoch少了

ConsistencyVC commented 1 year ago

对训练集中的语音,输出也是电流声吗,还是只是对测试用的语音输出是电流声。我记得你说原预训练模型对你的语音的输出是没有电流声的。可能是训练epoch多了,模型对这300条语音过拟合了。解决方法要么是减少训练epoch试一试,要么是把这300条语音混在大的语音数据集里。

tppqt commented 1 year ago

我试试看