FastMNMF cannot return source images from more than one microphone

[m-pexx]

The implementation of fastMNMF let you decide for which microphone the source images are returned. There is no possibility to return the source images for more than one microphone even if they are estimated from the algorithm.
I believe it would me more consistent with paper, if the function returns the source images for every microphone by default and let the selection to the user or as a side feature.

[fakufaku]

Hi @m-pexx Thanks for your very relevant suggestion. This is something I have been thinking about for a while. My thinking was that in most cases, regular users (i.e., not researchers) just want one separated signal per source. Then again, hard coding the reference microphone is not a great idea and could potentially lead to bad outcomes.

One idea I had was to use the proj_back parameter to pass the channel number on which to project. We could set the default to 0 to keep the current behavior and add an "all" keyword (or similar) to get all the images.

Another idea is to make a separate interface for the projection. If you have other suggestions, they are welcome. What is your use case ?

[m-pexx]

Hi @fakufaku I believe that an optional parameter specifying the indices of the microphones would do the job. You could leave 0 as a default while the user can specify a list of indices of the desired microphones.