Spectral Mask - Based Technique for Improving Generalized Sidelobe Canceller Beamformer's Evaluation

Noise suppression and understanding speech in a complex environment with the presence of numerous and dominant surrounding noise is very difficult for humans. To deal with this problem, microphone array, which exploit a priori information of designed geometry of microphone array, the direction of arrival (DOA) of interest signal or the properties of recorded situation, is used with aiming at preserving the desired speaker without speech distortion while reducing background noise. Generalized Sidelobe Canceller (GSC) beamformer has been installed in almost all acoustic equipment, hearing aids, mobile phones. Unfortunately, the performance of GSC beamformer is often corrupted in many scenarios, such as imprecise DOA, microphone mismatches or different microphone sensitivities. In this paper, a spectral mask - based suggested algorithm to reduce the speech leakage in the reference signal of GSC beamformer. The proposed method is evaluated for increasing the speech quality of dual - microphone systems.