Export Reviews, Discussions, Author Feedback and Meta-Reviews

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper presents a extension in the family of low-rank spectrogram-factorization models of sound decomposition that generalizes the latent factors over an arbitrary (potentially overcomplete) set of bases, meaning that decomposition can be performed on multiresolution (e.g. This is argued to help with source separation and denoising. It starts from Gaussian Composite Model of Fevotte et al, which models the variance of each cell of a spectrogram with a low-rank matrix factorization, but then expands the implicit convolution with the multiple gabor impulse responses corresponding to each row of the spectrogram to give a purely time-domain expression for what is being estimated, that can then be modified to work for different sets of impulse responses, including nonorthogonal ones. An EM (local optimum) analysis procedure is presented.