Study Of Speech Pitch Detection Algorithm Under Noise Environment

The speech pitch is a relatively important feature information in speech signal,which is mainly applied to speech synthesis system and voiceprint recognition system.At present,study of speech pitch detection algorithm has been the key points and difficultics of research,and the existing speech pitch detection algorithms are estimated,accuracy is better for pure speech signals,but the accuracy still needs to be improved for speech signal under noise environment.Therefore,this thesis used spectral subtraction method,adaptive subband filtering method,autocorrelation function and average amplitude difference function to solve the problem of low accuracy of speech pitch detection algorithm under noise environment.Firstly,this thesis analyzed the generation process and mathematical model of speech signal,researched the mathematical model of the pitch of speech signal,analyzed the characteristics of speech pitch,established the corresponding speech library,analyzed each process of preprocessing of the speech signal,and determined methods used for speech pre-emphasis,frame,windowing,and endpoint delection.The basic theory and algorithm of speech enhancement were studied,and the problem of spectral subtraction algorithm in the practical application of speech enhancement were pointed out,used the adaptive filtering algorithm to improve the spectral subtraction algorithm,and the post-adaptive filter spectral subtraction algorithm was proposed.The post-adaptive filter spectral subtraction algorithm not only solved the problem of "music noise" remaining in the process of using the spectral subtraction algorithm,byt also outputted a higher signal-to-noise ratio speech signal,at the same time,thec post-adaptive filter spectral subtraction algorithm is a applicable speech enhancement method under noise environment.This thesis simulated the autocorrelation function(ACF)and average amplitude difference function(AMDF)in fundamental frequency in speech detection algorithm,and pointed out the problem in their actual application.The correlation function method and the average amplitude difference function method are combined to improve the algorithm,the improved algorithm increased the peak value of the boundary point of the pitch period and the accuracy of the pitch period judgment,and further improve the accuracy of speech fundamental frequency detection algorithm.It retains the advantages of the autocorrelation function method and the average amplitude difference function method.B andpass filter,center clipping algorithm and smoothing filter algorithm are used in the algorithm flow,which reduces the interference of some formants,frequency doubling and half frequency,and wild pointsIn order to verify the effectiveness and accuracy of the improved speech pitch detection algorithm under noise environment,used the speech library data with noise to verify the speech pitch detection algorithm,counted error rate parameters PTE,UE,VE of speech pitch detection algorithm.For the speech library data set containing babble noise,the PTE of the improved speech pitch detection algorithm was reduced by 6.50%and 11.89%for the speech library data with pink noise,factory1 noise,and white noise,the PTE,UE and VE values of the improved speech pitch detection algorithm were less than the autocorrelation function method and the average amplitude difference function method.The results showed the improved speech pitch detection algorithm was more accurate than the autocorrelation function method and the average amplitude difference function method under noise environment,and the improved speech pitch detection algorithm could detect speech pitch of the speech signal in real environment.