Speech endpoint detection in fixed differential beamforming combined with modulation domain

Abstract

The actual speech signal is always in a complex environment, which affects the accuracy of speech endpoint detection. In order to improve the accuracy of speech endpoint detection under low Signal-To-Noise Ratio(SNR), a speech endpoint detection technology based on fixed differential beamforming and modulation domain is proposed. Firstly, two microphones of the first-order differential microphone arrays(DMAs) are used to collect the target speech and noise, and then a fixed differential beamformer is used to reduce the noise. Then, the phase compensated modulation domain spectral subtraction method is used to further eliminate the residual noise. Finally, the subband spectral entropy algorithm is used to detect the speech signal endpoint. The experimental results show that, compared with the existing methods, the proposed method has a higher accuracy of speech endpoint detection in the low SNR environment of -10∼5dB.