Title:A Complex Matrix Factorization approach to Joint Modeling of Magnitude and Phase for Source Separation

Abstract:Conventional NMF methods for source separation factorize the matrix of spectral magnitudes. Spectral Phase is not included in the decomposition process of these methods. However, phase of the speech mixture is generally used in reconstructing the target speech signal. This results in undesired traces of interfering sources in the target signal. In this paper the spectral phase is incorporated in the decomposition process itself. Additionally, the complex matrix factorization problem is reduced to an NMF problem using simple transformations. This results in effective separation of speech mixtures since both magnitude and phase are utilized jointly in the separation process. Improvement in source separation results are demonstrated using objective quality evaluations on the GRID corpus.