Unequal input array dimensions in create_dataset.py about transformertts

[ConProgramming]

Had the same error as you, my metadata.csv had a line where the transcription was empty... do a search in your editor for the delimiter then a new line. Remove the line with an empty transcription and it'll work.

[padmalcom]

It was the following entry in the metadata.csv:

menschenhasser_04_f000032|–, und eben da hat er so mancherlei durchblicken lassen.|-, und eben da hat er so mancherlei durchblicken lassen.

Removing the leading -, made it work.

[cfrancesco]

So it happens with both empty lines and leading , ?

@padmalcom do you know why the error arises? Seems like a phonemizer issue

[padmalcom]

I can't confirm the behavior for empty lines but with those starting with -,. I'm sure it is an issue with phoenemizer, but it can easily be avoided so I do not really see an issue here. Maybe implement a sanitizer step before creating the training data?

[cfrancesco]

Ok thanks. Yeah a cleaner is already in place, I'll try to understand the principle behind this and add it to it. (it also breaks when there is two punctuation and a space in between, like ", ,")