Research On Speech Enhancement Algorithm Based On Dual Microphone

With the development of the times,on the one hand,people’s living standards have improved and material life has become ample,the requirements for voice are getting higher and higher,microphone-based applications are becoming more and more extensive,and voice enhancement algorithms have become more and more important.People are also constantly pursuing higher sound quality and better listening enjoyment.Although microphone voice enhancement technology has been developed for decades,the performance of existing voice enhancement algorithms will be greatly reduced in a complex and strong noise environment.On the other hand,in the world’s population,more than 5% of people have suffered from varying degrees of hearing impairment.In the past ten years,there have been no major breakthroughs in the research and development of hearing aids and cochlear implants.Due to design limitations,The performance of hearing aids decreases when there are different types of background noise,and they cannot receive voice signals stably.One solution is to use two microphones to capture noisy voice data,perform complex calculations with a voice enhancement algorithm,and then transmit the enhanced voice to the hearing aid.The effect of traditional speech enhancement algorithms has gradually been unable to meet people’s needs.Based on this,this paper proposes two microphone speech enhancement algorithms.First,this article briefly introduces the research background of this subject and the current research status at home and abroad,then introduces several microphone algorithms,analyzes and compares their advantages and disadvantages,and then introduces microphone-based speech signal models and speech quality Evaluation criteria,focusing on the in-depth study of the super-Gaussian maximum posterior algorithm and the gain function based on the coherent algorithm,and proposed adding a "weighting" factor to the super-Gaussian coherent joint algorithm,which can be changed to control the weighted combination.So as to control the amount of noise suppression and speech distortion,experiments have proved that a better speech enhancement effect is obtained.Secondly,this article first proposes an improved dual-microphone gain function,and then systematically researches and analyzes the MVDR beamforming technology,studies the gain function based on the logarithmic spectrum amplitude estimation,and combines the two,MVDR beam The formed output signal is added to the gain function of the logarithmic spectrum amplitude estimation,which can minimize the mean square error of the logarithmic amplitude spectrum between the estimated speech and the real speech,thereby achieving the effect of speech enhancement.Finally,the pros and cons of the two microphone speech enhancement algorithms proposed in this paper are verified through experiments.Experiments show that the PESQ score（objective speech quality assessment）and CSII score（speech intelligibility）of the improved super-Gaussian coherent speech enhancement algorithm are higher than the logMMSE algorithm and the coherence gain algorithm.The second algorithm MVDR is proposed.It has the same superior performance as the logarithmic spectrum amplitude estimation joint algorithm,and the PESQ score and STOI score（short-term objective comprehension）are also higher than the other two algorithms.