Alexandre Défossez

Hello, Demucs author here, the reason is that PyTorch is not yet available for Python 3.10: https://github.com/pytorch/pytorch/issues/66424#issuecomment-985148728 Don't hesitate to tag me if you have future issues @boy1dr

We are considering the possibility of releasing this code under MIT, but it will take a bit of time to get the legal approval. At the moment we do not...

Thanks for the improvements :) Regarding grad computation I personally sur resampling on neural network outputs and in that case you must backpropagate the gradient to the network. For the...

Hey @xvdp sorry for the confusion with the gradient, I hadn't looked at the PR in details, indeed no gradient is needed for the kernel creation. Regarding the GCD approx,...

@akx due to internal change in pytorch you would need to rebase from main for the CI to pass, sorry about that!

Thanks a lot @ws-choi for the PR !! You just need to accept the CLA before I can merge this :)

You can try deactivating the fft based implementation (not sure anymore what is the name of the parameter), especially if you have an input that requires a gradient. It's tuned...

We use [Dora](https://github.com/facebookresearch/dora) for all of our experiments which typically calls `torch.cuda.set_device` at the beginning of the training with the proper device. That allows to use 'cuda' everywhere after without...

DNSMos definitely look interesting! Visqol is supposed to correlate to human perception but definitely has a lot of shortcomings and fail to capture the presence of artifact. While we are...

We do not aim for audio transparency with this codec. At the moment it is not so easy to get rid of the remaining artifacts even when allowing the model...