Problem with LJSpeechDataset

[melou-slr]

Hi @oleges1 ,
I have an error message when I launch 'train.py' with the LJSpeechDataset as config:
'LJSpeechDataset' object has no attribute '_walker'. Would you have any idea of ​​the problem, please?
Thanks a lot.

[oleges1]

Hi!
Seems like problem is in torchaudio version. I suggest you to check older versions: 0.7 or 0.6. Sorry for not stating this in requirements. It will definitely help.

Nevertheless, if you don't want to downgrade torchaudio, the problem is in deprecated internal item "_walker" in torchaudio.datasets.LJSPEECH. Now it's called "_flist" in newer versions, as can be seen on screenshot below. You can try to simply replace "_walker" to "_flist".

Link to the source code of torchaudio: https://pytorch.org/audio/stable/_modules/torchaudio/datasets/ljspeech.html#LJSPEECH

Thank you for reporting this!

[nguyennhuttin]

Hi I have a problem ( could be related to ljspeech as well any suggestion on how to fix the problem

File "C:\Users\nntin\anaconda3\lib\site-packages\torchaudio\datasets\ljspeech.py", line 72, in _parse_filesystem
self._flist = list(flist)
File "C:\Users\nntin\anaconda3\lib\encodings\cp1252.py", line 23, in decode
return codecs.charmap_decode(input,self.errors,decoding_table)[0]
UnicodeDecodeError: 'charmap' codec can't decode byte 0x9d in position 1936: character maps to

[oleges1]

Hi @nguyennhuttin , which version of torchaudio are you using?

[nguyennhuttin]

I am using

torch.version
'1.13.1'
torchaudio.version
'0.13.1'

[nguyennhuttin]

Hi, thank you but I think I manage to solve the issue by adding encoding='utf-8' to every with open csv file