Research On Speech Intelligibility Enhancement Based On Amplitude Spectrum Constraint

Speech communication is the most popular way in our daily life.It is always be used between people to people,people to machine.However noise interference is full of our life.Speech signals not only mixed with noise when it send out,but communication channels also generate white noise,this will resulting speech signals cannot accurately convey the original information.Therefore,it is necessary to use the speech enhancement technology to process the speech by noise,restore the original signal as accurately as possible,and improve the speech communication quality.The function of speech enhancement technology is to recover the original speech signal from the noisy speech and optimize the output speech quality and speech intelligibility.This thesis investigates the technology of speech enhancement based on amplitude spectrum constraints to improve speech quality and speech intelligibility.The main work of this thesis listed as follows:Firstly,this thesis focused on the enhancement of speech intelligibility,several types of typical speech enhancement algorithms for improving speech intelligibility in recent years have been summarized and analyzed.Briefly introduce four algorithms which is based on ideal binary masking,then explained the basic principles of each algorithm and analyzed the effect of enhanced speech intelligibility.Secondly,this thesis has carried out important research and analyzed the speech intelligibility enhancement based on speech amplitude spectrum constraints: 1?Aiming at the defect that the I+II constraint requires the prior information of the clean speech signal and cannot be directly applied,the amplitude spectrum constraint condition expressed by the gain function and the prior signal-to-noise ratio is obtained through detailed mathematical derivation;2?Combining the speech amplitude square spectrum estimation theory with the amplitude spectrum constraint algorithm,based on the complex Gaussian hypothesis model of speech and noise,two practically applicable speech intelligibility enhancement algorithms are proposed according to the maximum posterior and minimum mean square error criteria;3?The performance of the proposed algorithm is simulated and verified by using a variety of noise environments and speech and noise data.The output results of various speech enhancement algorithms are compared and analyzed in combination with evaluation indicators such as STOI,PESQ,and Seg SNR,which effectively demonstrates the excellent performance of the proposed algorithm in improving the intelligibility of noisy speech.Finally,summarized the basic framework and research content of the full text,and analyzed and forecasted the future research and development direction of speech enhancement.