Pariente Manuel

@gauss256 Did you go forward with the implementation of the DANet? I'm getting interesting in it and it would be great to work on it together, some things are blocking...

@zhr1201 IN DANet, if you use the anchored attractor points, you can implement it in "real-time", the main delay will be the one of the Fourier transform, so here 32...

Thanks for the issue. Probably by returning `y[tuple(keep)]`, do you want to submit a PR? Thanks !

It doesn't make sense to compute STOI for such short inputs, that's why. But it's true, the error can be improved, would you like to do it?

> I would be glad to implement it Super cool :tada: Yes, it's related to the window size used for the computation. Maybe it's mentioned in the paper.

Thank you, @hovavalon ! :pray: I'm linking the PR for reference: #24

I didn't check your files but are you sure they are completely synced between each other? If they are aligned, this results sounds weird indeed.

That's counter-intuitive.. Do you have Matlab by any chance? The code is unit tested but maybe something weird happens IDK..

Thanks a lot for running the tests in Matlab ! This is indeed a very interesting observation, @chtaal might have an explanation for it. I don't have any intuition as...

It is not invertible but the pseudo-inverse of the forward transform matrix is the way to go. See the back-propagable pseudo-inverse [here](https://pytorch.org/docs/stable/torch.html#torch.pinverse). Same goes for the mel-spectrogram by the way.