lixilinx / iva4cocktail Goto Github PK

I have a question, when the number of microphones is set to 4 or 5, the separation performance of the trained model is normal, but when it is set to 2, the separation performance is very poor

Hi,

In your paper or code, the density model parameters Ws is updated using Preconditioned SGD optimizer( this SGD algorithm is very interesting). If I want to use pytorch's optimization method, such as torch.optim.Adam( ), what should I do?
What I don’t understand is that there is still an updating of W_iva between density model and loss. How to do its backpropagation?

Hi,

just came across your work, quite interesting of replacing laplace distribution with neural nets' density model. But I did not get crystal clear about: how do we get the ground truth output to train these NN density models?