Applying the model to higher sampling rates

[eloimoliner]

Hi,

Thank you so much for sharing this great repository.

I was just wondering if you ever tried applying your model to unconditional audio generation using higher sampling rates (e.g. 44.1 kHz). Do you think there could be hope for it to work reasonably well, perhaps after modifying the architecture with some extra U-Net layers?

[albertfgu]

I haven't tried anything other than the two datasets documented in this repo, but I think there should be no problems with 44.1kHz. I'm guessing the architecture should work as-is, but you can certainly play with it; we didn't experiment much with the overall UNet structure so there might be room for improvement