jxzhanggg/nonparaSeq2seqVC_code

Alignment problem

Approximetal opened this issue · 2 comments

When I start training VCTK, it seems the alignment has something wrong.
step-28000-tts.pdf
step-28000-vc.pdf
I met the same problem when training tacotron2, and the author suggested to use pre-train model. Does this code has pre-train model? Or any other mothod to deal with this problem?

I haven't meet this alignment collapse problem. My code does not inclue a pre-trained model currently.
I suggest to exclude some long utterances in training data. Also, will the model recover
from this problem automatically if you keep training? If it will, I think you could still get normal results.

I haven't meet this alignment collapse problem. My code does not inclue a pre-trained model currently.
I suggest to exclude some long utterances in training data. Also, will the model recover
from this problem automatically if you keep training? If it will, I think you could still get normal results.

It won't recover I suppose, it should be correct at the very begining, otherwise nothing gonna change. I'll try to use other datasets. Thank you!