Noise reduction by using more reference signals in GSC beamformer

Abstract

Nowadays, the use of microphone array (MA) technology has been popular, due to its high spatial diversity, high directivity factor and the steered beampattern towards the specified sound location. MA beamforming uses the spatial information about the direction of arrival of a helpful signal, the designed geometry of MA distribution, the properties of surrounding noise, background noise to achieve noise reduction and speech enhancement at the same time. Generalized Sidelobe Canceller (GSC) beamformer is one of the most useful techniques for minimizing the total noise power while preserving the clean speech data at a certain direction without speech distortion. However, under complex recording scenarios, the overall MVDR’s evaluation often degraded and speech quality, speech intelligibility corrupted because of the moving head of speaker, the different microphone sensitivities, the error of sampling rate, the displacement of MA. In this article, we introduced a modified steering vector to increase the robustness of GSC beamformer in realistic recording situations. This approach is based on the preferred steering vector to achieve more reference signal to improve GSC beamformer’s performance in complex recording scenarios. The demonstrated experiments have confirmed the advantage of the proposed method in reducing musical noise, noise level at GSC beamformer’s output signal. The conducted experiment shows the improved signal-to-noise ratio (SNR) from 7.4 to 8.8 (dB). The suggested technique can be implemented in a multi-channel system for dealing other complicated problems, such as reverberation, speech recognition.