Robust source separation with differential microphone arrays and independent low-rank matrix analysis

Acoustic source separation has been an active and important area of research in the field of acoustic signal processing. This paper deals with this problem using small and compact differential microphone arrays (DMAs) so that the resulting technology can be used in a broad range of small devices in voice communication and human-machine interfaces. A straightforward way to achieve source separation with DMAs is through differential beamforming. Although it has frequency-invariant beampatterns and high directivity in comparison with other existing beamforming methods with the same number of sensors, differential beamforming with small DMAs has limited spatial gain, generally leading to insufficient separation performance. To circumvent this limitation, we propose in this work a method to combine differential beamforming with an independent vector analysis (IVA) based algorithm. Specifically, differential beamformers are designed and applied to separate sound sources from different directions. Then, differential beamformers' outputs are used as inputs for the independent low-rank matrix analysis (ILRMA) algorithm, a widely used IVA method for blind source separation. The advantage of this proposed method consists of at least three aspects: 1) improving the source separation performance, 2) helping deal with the permutation problem, and 3) helping improve the convergence of ILRMA.