Research On Speech Enhancement Algorithms Of Microphone Array

There are more and more electronic products with voice as the main interactive mode in our life,such as smart speakers,intelligent robots and so on.Adding more "wisdom" and "fun" to people's lives.However,there is a common problem of speech recognition that is accuracy reduction in noise environment for various types of "intelligent voice" application products,which are being promoted and used in the domestic and foreign markets.Therefore,it is necessary to use speech enhancement technology to denoise and improve the accuracy of recognition.In this paper,a set of speech enhancement algorithms based on microphone array is improved and implemented.by studying related technologies.Firstly,describing the mircrophone array technology and summarizing the main issues for current speech enhancement technology.Secondly,the algorithm of speech signal preprocessing,voice activity detection(VAD),direc-tion of Arrival(DOA)are studied.In this paper,a complex algorithm of voice activity detection based on multi-window spectral estimation spectral subtraction and energy-entropy ratio is used to solve the problem of "leak" detection in VAD with single parameter in low signal-to-noise ratio(SNR)environment.In the module of DOA,one method is improved by utilizing the energy difference between signal and noise bands.This method is based on the theory of beam directivity method,and the direction of Arrival by adding the weighting factor of the frequency band with higher signal-to-noise ratio(SNR).The experiment shows that this algorithm can effectively reduce the influence of noise frequency band and improve the accuracy of DOA.Next,the theoretical analysis of several adaptive beamforming algorithms for LMCV,GSC and MVDR is carried out.Based on the theory of minimum variance distortionless response(MVDR),a method of superdirective beamforming is improved and optimized through replacing the noise covariance matrix in the MVDR algorithm with diffuse noise field matrix,then the simulation test and result analysis.The test results show that this method can make the gain of the array output signal larger,the signal beamforming has better directivity,and has better enhancement effect on the actual speech signal.Finally,the design of the hardware application platform and main implementation process of the algorithm are described.The microphone array circuit used in signal acquisition and its characteristics are mainly analyzed.Then,based on the whole set of enhancement algorithms in this paper,the simulation test and multi-angle result analysis of the speech signal data collected from different sound source angles are carried out.In this paper,through the research and analysis of the characteristics about the speech signal noise environment,the important algorithms of speech enhancement,such as direction of Arrival and beamforming,are improved and optimized.The test results show that the improved algorithm achieved the design goal,it has the ability to reduce noise for the actual speech environment,and the robustness is improved,which has practical value in certain aspects.