Study On Adaptive Beamforming And Post-filtering Technology For Differential Microphone Arrays

In recent years,artificial intelligence has become a hot research topic in the technology field.As the interface of human-computer interaction,voice has become one of the best ways for the development of artificial intelligence.Microphone arrays,as a speech enhancement technology that efficiently suppresses noise,can highly improve the quality and intelligibility of speech in complex noise scenarios.Microphone arrays can be classified into additive microphone arrays and Differential Microphone Arrays(DMA)And both types of microphone arrays have their own advantages and disadvantages Among them,DMA has attracted wide attention due to its small size and maximum spatial directivity,but its poor robustness to system errors has seriously hindered its applications.This paper mainly studies the adaptive beamforming and post-filtering technology for the fourth-order differential microphone arrays,and alleviates the problems existing in the traditional method.The main work is as followsFirst,aiming at the poor robustness of differential microphone arrays in which the white noise is seriously amplified in low frequency bands,this paper improves the traditional DMA beamforming algorithms.By synthesizing two existing DMA fixed beamforming algorithms,a new beamformer has been developed.The improved DMA beamforming algorithm improves its White Noise Gain(WNG)in the low frequency band while ensuring its high directivity,and enhances the robustness of the algorithm under interference noiseSecondly,in view of the leakage of the desired signal of Blocking Matrix(BM)module in the traditional GSC algorithms,the GSC algorithm based on the dual BM module is proposed in this paper.The improved GSC structure contains two new BM modules,in which the improved multi-channel BM compensates the energy attenuation in low frequency bands of the traditional multi-channel BM through the equalization filter,thereby improving the noise reduction effect of Adaptive Noise Canceller(ANC)Another single-channel BM uses the improved DMA beamforming algorithm to achieve more accurate noise estimation,and then controls the update of adaptive filter coefficients in ANC by estimating the signal-to-interference ratio(SIR).Simulation results show that the proposed method can suppress noise more efficiently while reducing the distortion of the desired signalFinally,aiming at the shortcomings of GSC's weak ability to suppress non-coherent noise,this paper proposes a multi-channel post filter based on the noise power spectrum estimation of the BM module.This paper uses the improved BM module to accurately estimate the noise power spectrum,so as to calculate the gain function of the multi-channel Wiener filter to achieve the improvement of the quality of speech and recognition.Simulation results show that the improved post filter achieves better noise reduction and recognition results than the traditional post filter.