Design of Spectrogram-Consistency Regularization Term Dependent on Observation in Independent Low-Rank Matrix Analysis for Blind Source Separation

Independent low-rank matrix analysis (ILRMA) is the state-of-the-art technique for blind source separation under the overdetermined condition. Recently, some methods that focus on the spectrogram consistency of the separated signals to improve the separation performance have been proposed. One of such methods introduces into ILRMA the regularization term that directly induces the separated signals to be consistent. Although in the conventional method, the regularization term is designed to be independent of the observed signals, we design the regularization term to be dependent on the observed signals. For the obtained cost function, we derive a new update rule on the basis of the majorization-minimization algorithm with the auxiliary function that is an extension of that derived in the conventional method. Finally, a numerical experiment is conducted to verify the separation performance.