Research On Speech Enhancement Technology Of Microphone Array Based On GSC Structure

In daily life,the voice signal received by the microphone will be polluted by various noises.This phenomenon is inevitable.Therefore,the mixed speech signal usually improves the speech quality before processing,and the speech enhancement technology can be used to preprocess the mixed speech signal to obtain a purer speech signal.Compared with traditional single-channel microphone speech enhancement algorithms,microphone array speech enhancement technology can better deal with directional interference noise in complex noise environments and improve speech clarity.The main contents of this article are as follows:1.Describes several classic microphone array beamforming algorithms including GSC algorithm,and sets up simple and complex acoustic environments.The GSC beamforming speech enhancement algorithm and the fixed beamforming speech enhancement algorithm are in two situations respectively.Perform experimental analysis and comparison.It is concluded that the GSC algorithm has the inherent defects of poor incoherent noise suppression and signal self-cancellation.2.Aiming at the defect of the poor incoherent noise suppression of the GSC algorithm,an improved GSC-SS speech enhancement algorithm based on the spectral subtraction(SS)method is proposed,and the improved spectral subtraction is used as a postposition to perform the GSC algorithm.In cascade,power spectrum subtraction is performed on the output signal of the normalized improved GSC algorithm.The experimental results show that the algorithm can effectively improve the defect of the GSC algorithm's weak incoherent noise suppression ability,but it is easy to produce "music noise".3.Aiming at the deficiency of "music noise" in the GSC-SS algorithm,an improved GSC algorithm based on Kalman filtering is proposed.The Kalman filtering algorithm is used as a post-position,and the current value is estimated by using the previous state and the current state observation value.The voice signal after removing the directional interference noise is output to the cascaded Kalman algorithm.The residual background noise of the GSC output signal is estimated by the iterative minimum mean square error,thereby suppressing the incoherent noise and the thermal noise generated by the microphone array element.The experimental results show that the algorithm has obvious advantages in noise suppression compared with the traditional GSC algorithm and the post-spectral subtraction filtering algorithm,and the enhanced signal can retain more useful information.4.Aiming at the self-cancellation phenomenon of GSC in a complex noise environment,an improved GSC algorithm based on DNN is proposed.The algorithm replaces the lower branches in the GSC structure with a multi-objective DNN,and relies on the powerful data fitting ability of the DNN to improve the self-cancellation phenomenon.Experimental results show that compared with the traditional GSC algorithm,this algorithm has a good ability to adapt to complex noise environments.Compared with the traditional GSC algorithm and the other two algorithms mentioned in this article,the enhanced speech of this algorithm is closer to the target speech signal.