Christian J. Steinmetz

In addition to the recurrence, I have found vectorization of the compressor nontrivial in PyTorch. `torch.vmap` could potentially address that though.

The reason this currently doesn't work is that the desired behavior for `reduction="none"` when using the multi-resolution loss is a bit ambiguous. Since each `STFTLoss` will produce a different shape...

Perhaps the best way to achieve the desired behavior right now is to manage a set of `STFTLoss` instances yourself. That way you can define the behavior of how they...

`scale_invariance` is currently still a bit experimental, but shouldn't throw an error. Let me try to reproduce.

This is due to how things are defined in the original standard, they only specify behavior for 5.1 audio. I believe there has been more recent work to extend the...

Ah, I see there is an issue. The standard states: > The block diagram shows inputs for five main channels (left, centre, right, left surround and right surround); this allows...

So after taking a second look I think the current behavior may be the best option, however, there could be more documentation to describe the expected usage. What is important...

Thanks for raising this. I can't reproduce your code example due to some variables that aren't included. If I had to take some guesses, you may get bad results if...

Thanks for putting that together. I was able to validate that the solutions you are getting from IIRNet at in fact what the model is producing. It seems that the...

To properly test that filter coefficients are generated correctly we need to check our results against know good values. Here are some sources for the included filter designs: Filter |...