The Using of Spectral Mask to Enhance GSC Beamformer’s Performance

In the recording scenarios with annoying noise, interference, complex surrounding environment with different noise, the speech enhancement is an essential problem, which is recently studied in many decades. Generazlized sidelobe canceller (GSC) beamformer, which now widely applied in almost acoustic equipments for preserving the desired speech component with satisfactory quality while eliminating thirdparty talker, surrounding noise, non-directional interference, is an efficient microphone array (MA) beamformer. The realistic implementation of GSC beamformer often degraded due to many reasons, such as the different microphone sensitivities, the undetermined the direction of arrival of interest useful signal, the moving head of speaker during the conversation, the different microphone quality, the error of sampling rate or microphone mismatches. In this article, an effective spectral mask, which based on the observed coherence between two received array signal, is proposed for suppressing the speech leakage at GSC beamformer’s reference signal and enhancing the overall performance, the perceptual metric listener, obtained signal-to-noise ratio and speech intelligibility. The numerical experiment illustrated the advantage of using the suggested method in saving the target talker and increasing the speech enhancement.