Multispeaker Config about styletts2 HOT 4 CLOSED

You also need to change the decoder type to HiFiGAN and make sure the data file contains the speaker ids for the second stage training. I’m currently testing the code on LibriTTS and it seems to work for the first stage, though it’s not done yet.

The code is exactly same as single speakers, the only difference is that multispeaker models use HiFiGAN decoder and the style diffuser takes one extra speaker embedding as the input.

from styletts2.

I found a few bugs in the multispeaker settings. I have fixed them, but there might still be some other problems. I'm still testing everything so it may take another one to two weeks.

from styletts2.

I found a few other bugs in the second stage as I have finished the first stage training in evaluations. Hopefully I’ll finish the LibriTTS model by next week.

from styletts2.

I think I have fixed all the bugs in the code for training multispeaker models. I just need to wait for the last phase of second stage to be done for the LibriTTS model.

from styletts2.