The Performance of GC-AuxIVA-ISS Method in a Realistic Environment

In this paper, we successfully reproduced the geometrically constrained independent vector analysis with the iterative source steering(GC-AuxIVA-ISS) and tested its performance under simulation and realistic conditions. GC-AuxIVA-ISS is a method that combines three important parts: I. the well known blind source separation method AuxIVA; II. beamforming-based geometrical constraints, which are defined using the spatial information of the sources; III. the ISS method, which does not require matrix inversion, achieves a lower computational complexity per iteration; resulting in the algorithm being faster and more stable than AuxIVA. This new method allows us to achieve distinguished separation performance and be able to obtain the target speech at the desired output channel. The experimental results in simulation and realistic revealed that this method has higher source separation performance and accurate ability of output channel order controlling. However, in a noisy reverberant condition, a single BSS method can hardly deal with such a difficult situation, the separated source is neither clean nor clear and the output sequence control is hard to apply. Some assisted dereverberation and denoise approach are needed in future realistic applications.