Bayesian two source modeling for separation of N sources from stereo signals

We consider an enhancement to the DUET sound source separation system (Yilmaz, O. and Rickard, S., IEEE Trans. Sig. Process., 2002), which allowed for the separation of N localized sparse sources given stereo mixture signals. Specifically, we expand the system and the related delay and scale subtraction scoring (DASSS) (Master, A.S., "Sound source separation of n sources from stereo signals via fitting to n models each lacking one source", Tech. Rep., CCRMA, Stanford University, 2003) to consider cases when two sources, rather than one, are active at the same point in STFT time-frequency space. We begin with a review of the DUET system and its sparsity and independence assumptions. We then consider how the DUET system and DASSS respond when faced with two active sources, and use this information in a Bayesian context to score the probability that two particular sources are active. We conclude with a musical example illustrating the benefit of our approach.