An Improved MVDR Beamformer's Performance by Using Dual - Microphone System

Microphone array (MA) technology has been installed in various types of acoustic equipment,such as: surveillance devices, hands-free mobile phones, teleconference systems, hearing aids, voice - controlled devices. MA uses the a priori spatial information about geometry's configuration, the properties of environment to form a spatial filter, which suppresses noise and speech enhancement at the same time. Minimum Variance Distortionless Response (MVDR) beamformers are the most useful and widely commonly implemented for speech enhancement applications. The original MVDR beamformer is significantly sensitive to the estimation of direction of arrival (DoA) of interest, clean speech component, microphone mismatches, the different microphone sen-sitivities, the imprecise MA distribution and undetermined characteristics of the surrounding environment. To improve the robustness of MVDR to extract the desired target speaker without speech distortion while removing background noise, in this article, the authors proposed an increased MVDR beamformer's evaluation by using an efficient spectral mask. The illustrated experiment has confirmed the effectiveness of suggested technique by enhancing the speech quality, speech intelligibility and perceptual listener in the term of the signal-to-noise ratio (SNR) from 5.7 to 8.1 dB and reduce speech distortion to 9.0 dB in the realistic recording scenario. As a result, the proposed technique has the capability of enhancing MVDR beamformer's evaluation and it can be integrated into a multi-channel system to solve several complicated problems.