Speech Enhancement With Microphone Array Based On Spectrum Compensation

Speech is an indispensable way for people to transmit information.However,the complex noise seriously pollutes the speech signal.Therefore,how to effectively eliminate the noise of noisy speech signal is very important to improve the quality of speech.Speech enhancement with microphone array algorithm has better spatial selectivity and anti-jamming ability,which can enhance the target speech in specified direction,and effectively filter out environmental noise.However,It's very limited to suppress the incoherent noise,and the phase of the noisy speech remains unchanged when enhanced,and some noise components still remain after filtering.In view of the limitations of traditional speech enhancement with microphone array,this paper proposes a speech enhancement microphone array algorithm based on spectrum compensation.The main work of this paper is as follows:Firstly,the theory of microphone speech signal model,research background and significance,research history and current situation at home and abroad,microphone array signal processing model,basic features of speech signal and speech enhancement technology are introduced.Secondly,aiming at the problem of inadequate suppression of incoherent noise by traditional adaptive beamforming,an improved post-filtering(AW-MVDR)speech enhancement algorithm based on minimum variance distortionless response is proposed.The algorithm uses the minimum mean square error criterion to update the weighting factor of the post-filter adaptively,so as to determine the optimal spectral gain function,and then designs the subband filter to further improve the filtering accuracy.Experiments show that the PESQ score of this algorithm is improved by 0.296 points on average and 12.1% compared with the MVDR beamforming under the condition of Gaussian white noise,which show that The denoising effect is effectively improved.Thirdly,aiming at the problem that traditional adaptive beamforming keeps the phase spectrum information of noisy speech unchanged,a speech enhancement algorithm based on S-PSC is proposed,which modifies the phase spectrum compensation function according to the input signal-to-noise ratio of each frame.Because the accuracy of noise estimation has a great influence on the performance of phase spectrum compensation,an improved time recursive average algorithm based on the existence probability of speech is proposed to estimate the noise power and further improve the performance of phase spectrum compensation.Experiments show that the PESQ score of S-PSC algorithm in this paper is improved by 0.143 points on average and 7.9% compared with the traditional phase spectrum compensation under the condition of Gaussian white noise,and obtain better speech enhancement performance.Fourthly,This paper combines the advantages of amplitude and phase compensation.Thus,a speech enhancement with microphone array algorithm based on spectrum compensation is proposed.Firstly,the AW-MVDR post-filtering algorithm is used to estimate the amplitude spectrum of pure speech,and the S-PSC algorithm is used to compensate the phase spectrum and then reconstruct the spectrum information of the target speech signal in real time.The experimental results show that the PESQ score of the algorithm is improved by 0.422 points and 17.3% compared with the traditional MVDR under the condition of Gaussian white noise.It can be seen that the proposed algorithm can effectively suppress the noise components,and the output quality is significantly improved.