Frequency domain blind source separation for robot audition using a parameterized sparsity criterion

In this paper, we introduce a modified l_p norm blind source separation criterion based on the source sparsity in the time-frequency domain. We study the effect of making the sparsity constraint harder through the optimization process, making the parameter p of the l_p norm vary from 1 to nearly 0 according to a sigmoid function. The sigmoid introduces a smooth l_p norm variation which avoids the divergence of the algorithm. We compared this algorithm to the regular l₁ norm minimization and an ICA based one and we obtained promising results.