should resampling be moved to before random cropping here? currently it seems to result in batches which aren't multiples of the latent downsampling factor when converting 44.1/48k.
I notice it also assumes that datasets processed at 48k with older versions of RAVE (where the sample rate wasn't stored as metadata) are 44.1 and tries to convert them, which is confusing for anyone who has been using 48k and upgrades. maybe it should do no conversion by default if the data same rate isn't known, and print a warning? I did it like this on my fork:
https://github.com/victor-shepardson/RAVE/blob/ff2218369f1589b06587bb58f37a609dc483d464/rave/dataset.py#L336
https://github.com/acids-ircam/RAVE/blob/8b250310fecfe61a6d9d53e8e5551851f4638d35/rave/dataset.py#L239
should resampling be moved to before random cropping here? currently it seems to result in batches which aren't multiples of the latent downsampling factor when converting 44.1/48k.
I notice it also assumes that datasets processed at 48k with older versions of RAVE (where the sample rate wasn't stored as metadata) are 44.1 and tries to convert them, which is confusing for anyone who has been using 48k and upgrades. maybe it should do no conversion by default if the data same rate isn't known, and print a warning? I did it like this on my fork: https://github.com/victor-shepardson/RAVE/blob/ff2218369f1589b06587bb58f37a609dc483d464/rave/dataset.py#L336