Open freddy5566 opened 2 years ago
hi, thanks for your suggestion. this makes a lot of sense. We will update to fix it
Sorry for the late reply, thank you for your work. Just out of curiosity, when will it be updated?
hmm, it looks like that torchaudio has a bug when loading 16 bit / 32 bit.
The current model depends on numpy int16 for feature extraction, but torchaudio's loading is float by default and somehow it fails to load int16 even I specified its normalization config (it loads as int32 which is overflowing int16, so it cannot be casted. forcing cast corrupt the results).
I am currently trying to upgrade to a new version removing most numpy dependency and using all torch feature including the torchaudio, so I guess I can only fix it when releasing the new model.
okay, thanks for your help.
Hi,
It seems like the wave package does not support 32-bit floating encoding. Here is the error message:
Could we try to use torchaudio instead of the wave to open files?
Thank you