Research On Speech Separation Algorithm Based On Fuzzy Clustering And Deep Learning

As a modern signal processing technology,underdetermined blind source separation has been widely used in the fields of speech,biomedicine,fault detection and so on.With the development of intelligent speech system,speech separation has an important impact on speech recognition and speech enhancement.In practical application,because the number of channels receiving speech is generally less than that of mixed speech,the research on underdetermined blind source separation technology of speech signal is of great significance to promote the development of speech separation technology.This paper focuses on the clustering problem in underdetermined speech blind source separation.For small data sets,an improved fuzzy clustering algorithm is designed to solve the mixing matrix estimation in sparse component analysis(SCA),so as to provide a more accurate mixing matrix for source signal separation;For large data sets,fuzzy clustering algorithm and deep learning technology are combined to separate underdetermined speech signals.The main work of this paper is as follows:(1)In the "two-step" framework,the traditional fuzzy clustering algorithm needs to specify the number of clusters in advance when estimating the mixing matrix,and it is easy to fall into local optimization.In order to solve this problem,this paper designs an improved fuzzy clustering algorithm based on density peak,which can not only improve the estimation accuracy of mixing matrix,but also automatically obtain the number of source signals;At the same time,this paper also designs an improved particle swarm optimization algorithm based on density peak,and compares it with the improved fuzzy clustering algorithm.The experimental results show that the proposed algorithm has higher estimation accuracy of mixing matrix than other algorithms.(2)Due to the complex structure of large-scale data sets and the unclear correlation of internal data,in order to solve the problem that the deep features can not be fully mined and the real-time performance of calculation can not be guaranteed in the traditional "two-step" processing of large-scale speech data sets,this paper applies fuzzy clustering to the deep learning framework,and designs a deep fuzzy clustering model to separate speech from large-scale speech data sets,The experimental results show the effectiveness of the proposed algorithm.(3)In order to explore the influence of deep clustering models based on different clustering algorithms on the final speech separation effect under different network structures,this paper trains different neural network models by changing the number of hidden layers of the network,and then carries out speech separation.The experimental simulation analysis shows that when the network structure is small,selecting the deep clustering model based on fuzzy clustering proposed in this paper can improve the accuracy of speech separation,When the network structure is large,the precision of the proposed deep clustering model based on fuzzy clustering is not much different from that of the traditional deep clustering model.