The Research And Realization Of Speech Enhancement Based On Noise Spectrum Estimation

Speech signals are used in many fields because of their advantages of direct,efficient and convenient.However,while using voice to convey information,the noise in the actual environment will inevitably affect it.Noise not only affects the comfort of speech receivers but also affects the performance of various speech processing applications.The purpose of Speech enhancement is to remove the noise part from the noisy speech as much as possible,and restore the pure speech signal,so as to reduce the influence of speech clearness and intelligibility.It is very important to estimate the characteristics of noise in speech enhancement algorithm.If noise is estimated to be too small,the noise will not be removed completely.If noise is overestimated,which will damage the pure speech and cause speech distortion.Therefore,it is very important to research noise estimation algorithm to improve the accuracy of noise estimation.In this paper,several representative noise estimation algorithms are introduced and compared.Then,based on the Improved Minima Controlled Recursive algorithm(IMCRA)algorithm,By analyzing the time delay of the minimum search in the algorithm,the improved noise estimation algorithm is proposed.The improved noise estimation algorithm can improve the accuracy of noise estimation and improve the quality of speech enhancement.On the basis of improved noise estimation algorithm,speech enhancement based on improved algorithm is applied to speech recognition,and a speech recognition system is designed and developed.The main works of this paper are as follows:(1)When the noise levels goes up,IMCRA noise estimation algorithm has the problem of tracking delays,and the estimate is not accurate.Concerning this issue,using the continuous minimum tracking algorithm to replace minimum statistical algorithms in IMCRA algorithm,so that the minimum tracking window are not binding and reduce time delay;In addition,a bias compensation function model based on the probability of speech is proposed to compensate the deviation of each band.An improved algorithm for noise estimation is proposed to improve the IMCRA algorithm for the above two points.And then,contrast experiments are made on the improved algorithm and the IMCRA algorithm.The experimental results show that compared with the original algorithm,the improved algorithm can effectively improve the quality of the enhanced voice in both smooth and non-stationary noise environments.(2)Finally,the speech enhancement system based on improved noise estimation algorithm is applied to the speech recognition scenes,a speech recognition system is designed and developed by using JAVA Swing programming.This system implements functions as real-time recording voice,enhancing the voice,voice recognition,and so on.This illustrates the usefulness of the speech enhancement techniques.