Research On Speech Enhancement Algorithm Based On Sparse Coding

Speech is the most convenient and fastest way for communication between people and people or people and machine.With the rapid development of information technology,the use of digital methods to process speech signals is one of the most fundamental and important components of the overall digital communication system.When people acquire speech signals,they will inevitably be disturbed by various noises from the surrounding environment or the internal equipment,which will degrade the performance of the speech obtained by the receiver.The purpose of speech enhancement is to extract the original clean speech signal from the noisy speech signal as much as possible,so as to improve the quality and intelligibility of the speech signal.Speech enhancement plays an important role in reducing noise interference,improving speech quality and intelligibility.It is one of the urgent problems to be solved in the practical application of digital speech signal processing.This paper mainly studies single-channel speech enhancement method.The speech enhancement algorithm is studied and summarized from two aspects.On one hand,it is a traditional classical speech enhancement algorithm,on the other hand,it is an emerging speech enhancement algorithm.Based on the existing speech enhancement algorithms,this paper proposes two improved speech enhancement algorithms based on sparse theory to solve the key technical problems of of sparse reconstruction algorithm in reconstructing pure speech,and simulation experiments are carried out.The main research work of this paper is as follows:Firstly,this paper systematically introduces the relevant knowledge of speech,studies speech enhancement algorithms,classifies them,and introduces in detail the theory of some classical speech enhancement algorithms.Secondly,the sparse representation model of speech signal is introduced in detail in this paper.And the sparse reconstruction and dictionary learning algorithms are systematically studied.This paper is focused on the study of traditional greedy sparse reconstruction algorithm,such as Orthogonal Matching Pursuit(OMP)algorithm.An improved OMP algorithm is proposed for the problem of OMP algorithm's ability to reconstruct speech signals.Firstly,a threshold is added to the OMP algorithm,and the reconstructed speech frame is adaptively selected by using the sparse threshold where an appropriate clean speech frame is selected from the noisy speech,thereby obtaining the best speech signal reconstruction performance,which is called improved speech enhancement performance for the OMP algorithm.At the same time,to improve the operation speed of the OMP algorithm,the matrix transformation is used in the operation process to reduce the steps and dimensions of the corresponding calculation,thereby improving the operation speed,which is called the improvement of the operation speed of the OMP algorithm.Based on the improved OMP algorithm above,a speech enhancement method based on K-SVD and improved OMP algorithm is proposed.The experimental results show that the proposed algorithm improves the computational speed of the algorithm while improving the speech enhancement performance.Finally,using the characteristics of the speech signal,based on the existing classification of the unvoiced and voiced speech,the zero crossing rate and the average energy are combined to carry out the classification of the unvoiced and voiced speeches,and then with the unvoiced and voiced classification method combined with the sparse representation,a sparse adaptive speech enhancement algorithm based on unvoiced and voiced sound classification is proposed.Firstly,the speech that needs to be trained is classified into unvoiced and voiced speech,then the corresponding unvoiced and voiced sub-dictionary is obtained by K-SVD dictionary training algorithm,and then the unvoiced and voiced sub-dictionaries are combined into a dictionary,and the improved OMP algorithm above is used to perform sparse decomposition and reconstruction to achieve the purpose of denoising.Compared with the above-mentioned speech enhancement algorithm based on the improved OMP algorithm,this algorithm also achieved good experimental results.