Improved Single Channel Speech Enhancement Algorithm Based On Compensated Phase Spectrum

Sending information through speech is the most important,effective and convenient form of common information exchange in human life.In this information age,the amount of information contained in speech plays a vital role.However,since there are all kinds of noise around in real life all the time,speech enhancement technology has a decisive position.Speech enhancement technology has been a significant research topic in speech signal processing at home and abroad,and a lot of research work has been done.In this regard,before introducing the improved single-channel speech enhancement algorithm,this article first elaborated from the aspects of speech enhancement research background and significance,research status and development trends,and the difficulty analysis of speech enhancement research.In order to make the article clearer,some theoretical foundations of speech enhancement are also described,such as: speech signal characteristics,noise characteristics,preprocessing,two basic speech enhancement algorithms,a band division method for comparison and four subjective and objective evaluation methods.In the final synthesis process of the classic spectral subtraction in the past,the original noisy speech phase spectrum and the speech amplitude spectrum were combined.However,the latest psychoacoustic research shows that the speech phase spectrum also has a greater degree of intelligibility contact.Therefore,the traditional phase spectrum compensation algorithm proposes a way to correct the phase spectrum,which is additive phase spectrum compensation to the complex spectrum of the noisy speech.This paper proposes an improved phase spectrum speech enhancement algorithm using the equivalent rectangular bandwidth(ERB)scale division.This algorithm is improved on the basis of the traditional phase spectrum compensation algorithm,and the noise estimation amplitude in the algorithm is calculated by multiband spectral subtraction using ERB scale.The improved compensation phase spectrum is an additive calculation of the preliminary enhanced speech complex spectrum.At the same time,the improved algorithm also applies ERB scale division to the speech algorithm that compensates the preliminary enhancement in the phase spectrum.In this paper,the compensated phase spectrum is additively calculated on the complexspeech spectrum after the initial enhancement of multiband spectral subtraction based on ERB scale division.Judging from the results of objective evaluation and the subjective evaluation of intuitive auditory perception after using Matlab to build the simulation environment,the algorithms proposed in this paper have obvious advantages over the four contrasting speech enhancement algorithms.In addition,this paper also improves the compensation constant in the traditional phase spectrum compensation function.Due to the improvement of the traditional compensation phase spectrum speech enhancement algorithm,the compensation constants derived from experimental experience in the previous traditional algorithms are no longer applicable,and the constants cannot be used to flexibly compensate the background noise in different situations.Therefore,the originally fixed compensation constant is improved to a compensation factor that changes according to the change value of the segmented signal-to-noise ratio,which makes up for the problem that the noisy speech phase spectrum cannot be flexibly compensated.The improved algorithm in this paper can flexibly compensate the phase spectrum of noisy speech according to the different changes of noise in each frequency band,which is combined with the previously obtained improved compensation phase spectrum speech enhancement algorithm using ERB scale to divide the critical band.From the objectively evaluated simulation results and the results of subjective scoring,it is shown that under the conditions of specific signal-to-noise ratio and specific noise,the processing effect of the algorithm proposed to add compensation factors has been significantly improved,and in other cases it is not worse than the original algorithm.