Fabian-Robert Stöter

nnAudios implmentation seems to be onnx exportable. We might want to check the differences... https://github.com/KinWaiCheuk/nnAudio/issues/23#issuecomment-768954731

@mpariente next issue, to address the decoder, torch.fold is not supported.... https://github.com/asteroid-team/asteroid-filterbanks/blob/351029209556bdccf6d27edbebfc6c8988c663ff/asteroid_filterbanks/torch_stft_fb.py#L172-L176 whats good replacement?

> Are you sure about that? > Replacing it will be very cumbersome seems so: https://github.com/pytorch/pytorch/issues/41423

@mpariente @jonashaag works with torch 12 and opset > 11 now! Should I still ad some tests?

@YuanGongND 👆

good, idea. Also since e.g. there is not that much of pytorch specific things when using it together with pescador, but it might still help others to start. Thats why...

we put the usage of pescador in combination with pytorch on hold for now until a non-index based dataset class is implemented. Of course, one could use pescador in combination...

> I have high hopes in a new Iteratable dataset that is currently developed by the pytorch folks. So, stay tuned... its merged and available in the 1.2 nighly built...

@cjacoby thanks for your input. I thought about this a bit more and tried many things. In the end I agree: You loose a lot of flexibility by using the...

I now had some more time to dig into pytorch and got nice results together with pescador. @bmcfee: > haven't looked at any of the other packages you mentioned --...