CJay

NaN during training when using own dataset

1

While fine-tuning works as expected, doing regular training with a dataset that isn't LJSpeech would eventually cause a NaN loss at some point. The culprit appears to be the following...

Inference produces identical output for different speakers

1

I'm not sure if this is a bug or a training issue, but my trained models always produce the same output for both speakers: ``` [hannes@chonker voice-conversion]$ python inference.py -f...