Research On Speech Signal Enhancement Algorithm Based On MATLAB

Voice enhancement technology is an indispensable and important part of voice communication and information transmission.It plays an irreplaceable role in today's life and production.The development of speech enhancement technology is changing with each passing day,and various enhancement methods have sprung up.Their research and development has always been an important issue.Today,traditional single-channel enhancement algorithms,such as MMSE,Wiener filtering,etc.,have already had relatively mature applications in information communication.For the constant pursuit of communication quality,the enhanced voice that single-channel speech enhancement can provide has gradually failed to meet people's requirements for speech quality.Therefore,multi-channel enhancement methods have entered the sight of people,especially the emergence of microphone arrays.It has better performance in noise reduction,distortion reduction and other aspects.This article will discuss the speech enhancement technology from the following two aspects.(1)The MMSE-LSA algorithm is discussed in terms of single-channel speech enhancement and its performance is analyzed.The gain function in the original algorithm is mainly determined by the a priori SNR,which results in latency and noise residuals in the enhanced speech.The problem is that the algorithm uses the speech signals of the previous frame and the current frame.Based on the relationship between a priori signal-to-noise ratio and a posteriori signal-to-noise ratio,a method based on a posteriori SNR is used to estimate the a priori signal-to-noise ratio.The delay problem,which gives the best estimate of noise,is determined by the fact that the smoothing factor ? is no longer completely dependent on the a priori signal to noise ratio,and the smoothness factor ? can be dynamically determined by using inter-frame smoothing to introduce the a posteriori signal to noise ratio.The signal-to-noise ratio is tracked to adapt to changes,and the residual music noise is suppressed to reduce the signal distortion as much as possible to improve the quality of speech.Through simulation experiments in MATLAB environment,it is shown that this method has better performance than other enhancement methods.(2)Discuss the existing multi-channel speech enhancement methods,and mainly carry out further research on sub-space-based multi-channel enhancement methods,simulate their performance through MATLAB,and analyze their advantages and disadvantages.It proposes improvements to its performance.In the original algorithm,the distortion of the enhanced speech and the residual noise are not resolved to achieve the desired effect.This paper proposes a method of iterating in a generalized subspace,and the characteristics of thecovariance matrix of the speech through repeated one-dimensional iterations.The values??and eigenvectors are estimated iteratively.At the same time,the value of the parameter ?in the gain matrix is ??improved.The improved DD algorithm is used to estimate the signal-to-noise ratio optimally to ensure the optimal value of the parameter ?.With the best gain matrix,the enhanced speech signal reduces the residual noise while simultaneously reducing the distortion of the speech and has better speech quality.In the MATLAB environment,a large number of experimental verification,the algorithm in the enhancement process,do not have to whiten the non-white noise pretreatment,and for each noise has a better processing effect,can also be achieved in a complex environment Good enhancement effect.