A blind source separation criterion where approximate disjointness meets independent component analysis

This paper proposes a sparseness-based blind source separation (BSS) method. In contrast to conventional approaches, we exploit the sparseness property and the ensuing approximate disjointness of the competing audio signals when represented in the short time Fourier transform domain to determine the linear separating matrix such that its outputs are maximally disjoint. By doing so, we deduce an iterative gradient descent to estimate the optimal separation matrix. Interestingly, the resulting optimization problem is shown to have strong links with independent component analysis using higher order statistics, and shares some similarity with non-stationarity-based BSS. The purpose of the proposed study is to provide some insight into the connection between the seemingly different sparseness and independence-based BSS criteria.