Single Channel Signal Separation Using MAP-based Subspace Decomposition

Introduction: Extracting multiple source signals from a single channel mixture is a challenging research field with numerous applications. Conventional methods are mostly based on splitting mixtures observed as a single stream into different acoustic objects, by building an active scene analysis system for the acoustic events that occur simultaneously in the same spectro-temporal regions. Recently Roweis presented a refiltering technique to estimate time-varying masking filters that localize sound streams in a spectro-temporal region [1]. In his work, sources are supposedly disjoint in the spectrogram and a “mask” whose value is binary, 0 or 1, exclusively divides the mixed streams completely. Our work, while motivated by the concept of spectral masking, is free of the assumption that the spectrograms should be disjoint. The main novelty of the proposed method is that the masking filters can have any real value in [0, 1], and that the filtering is done in the more discriminative, statistically independent subspaces obtained by independent component analysis (ICA). The