HighwayConvBlockGenerated class return size miss match error about multilingual_text_to_speech

[Tomiinek]

Hello,

Could you print shapes of x,h2, and most importantly self._groups after chunking? h should be two times larger than x, so after chunking to odd and even things everything should be ok.

I guess that the problem is caused by invalid self._group which should be the same as the number of languages in your batch when using the perfect sampler (the inputs for all languages are encoded in parallel using the grouped convolutions to avoid a need for multiple encoder forward passes).

[Thien223]

I have 7 languages in dataset, so self._group variable is 7. Is that true?

As you said that the self._group is the number of language in the batch only. But I have traced in the code and think that self._group is language number from hp.language_number (not in the batch). Is there something i have missed?

Thank you so much, sir.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.