Research On Speech Emotion Recognition Technology Based On Compressed Sensing

With the rapid development of information technology and network communication,people not only concern about the accurate acquisition of the signal,but also hope to understand the signal more deeply,which is a revolution in the field of signals.As the most important tool for human communication and human-computer interaction,recognition and emotional analysis of speech signal is increasingly important in the field of high-tech,such as Internet,communications,artificial intelligence and so on.So many experts accomplished lots of works of recognition and emotional analysis of speech signal.In these research directions,the research of speech content recognition is increasingly mature and has already been available for commercial use.Recently,although many research results on speech emotion recognition have been published and has achieved great development on speech emotion recognition,the research is still in the initial stage,and has not yet formed a set of widely accepted,systematic theories and research methods.In addition,because most of the real-life speech signals contain noise,and the traditional speech emotion recognition algorithms are mostly applied to pure voice,finding a robust speech emotion recognition technology is extremely urgent.Compression sensing is a new sampling technique proposed in recent years,and the sparse representation recognition algorithm based on compression sensing shows excellent recognition performance in image and speech recognition.And emotional speech signal has a good sparsity in wavelet transform domain.Therefore,the application of compression sensing technology to speech emotion recognition has strong theoretical basis.In addition,the compression sensing reconstruction algorithm has a certain inhibiting effect on the noise,and there are less researches on speech emotion recognition in the noisy environment.Based on the above analysis,on the basis of the existing research,we combine the compression sensing technology with the speech emotion recognition to make it possible to further enhance the performance of speech emotion recognition system,which has great application value on both theoretical and practical.This paper carries out the research on speech emotion recognition based on compression sensing for pure speech and speech in the noisy environment.The specific contents are as follows:1)Through the study of speech emotion recognition system theory and compression sensing algorithm,the research applies compression sensing to speech emotion recognition,and proposes a speech emotion recognition algorithm based on compression sensing.Simulation results show that the speech emotion recognition system based on compression sensing can achieve better recognition results compared with the classical GMM speech emotion recognition algorithm.The experimental results verify the correctness of the application of compression sensing to speech emotion recognition and the combination broadens the thought of speech emotion recognition algorithm.2)In order to further improve the recognition performance of speech emotion recognition system for pure speech and speech in the noisy environment,this paper proposes a speech emotion recognition system with compression sensing noise reduction algorithm.Firstly,the sparse reconstruction of the noisy speech is carried out by using the compression sensing technique,and the acoustic features of the sparse reconstructed signal are extracted.Finally,the features are input to the traditional GMM classifier.The simulation results show that the proposed method can achieve a large recognition gain and recognition rate increased by 5 to 32 percentage points compared with the emotion recognition system which does not use the compression sensing speech suppression.3)In order to further improve the performance of speech emotion recognition system,,this paper proposes a speech emotion recognition algorithm based on sparse Bayesian learning,dealing with the defects of sparse solution of compression sensing Firstly,the features of the speech signal in noisy environment are extracted,and the features are reconstructed by the sparse Bayesian learning.Then,the reconstructed feature vector is used to calculate the reconstructed distance from the emotion codebook.The simulation results show that the sparse Bayesian learning algorithm can be closer to the solution of 10-norm,and it has smaller reconstruction error compared with FOCUSS and BP.Experiments show that the application of sparse Bayesian learning algorithm to the speech emotion recognition system improves the performance of the recognition system and shows the application potential of sparse Bayesian learning in the field of speech emotion recognition.In the end of this paper,a conclusion is made to summary all the research and achievements of the paper and the future research work is prospected.