Research On Speech Emotion Recognition Based On Kernel Function

As an important branch of emotion computing, emotion recognition has aroused extensive concern of researchers in recent years. Speech is not only one of the most important way to communicate, but also carries the speaker’s emotional information. Speech emotion recognition technology is able to make the computer to recognize speakers’ emotional states, so it helps to achieve a more harmonious human-computer interaction and has wide application prospects. This paper focuses on speech emotion recognition methods based on kernel which helps traditional speech emotion recognition methods to solve nonlinear problems. To further improve the accuracy of speech emotion recognition, this paper also proposes some improved methods. The main work and contributions of this paper are as follows:(1) This paper outlines the research background and significance of speech emotion recognition, and summarizes the research of the definition of emotion, emotional database, emotional features, feature dimension reduction methods and emotion classifiers.(2) An emotional speech database has been established for experiments and it contains emotional speech in five emotional states:happy, fear, anger, sadness and neutral. After preprocessing the voice signals, emotional features such as speech speed, energy, the pitch frequency, MFCC are extracted from the signals. All the work is the base of the speech emotion recognition.(3) An algorithm combined kernel C-means clustering with kernel K nearest neighbor classification is proposed for speech emotion recognition. This algorithm not only uses the kernel method to improve nonlinear processing capability of classifier, but also improves the classification speed and overcomes the traditional shortcomings of kernel K nearest neighbor classification. In order to further improve the recognition accuracy,this paper also applies the fuzzy sets theory in the algorithm. It gets more accurate clustering results and constructs a membership function which makes test samples belong to each emotional category with different degree. Final results show that the algorithm is more accurate.(4) Kernel sparse representation classification algorithm is applied in speech emotion recognition. This algorithm uses kernel mapping mechanism to improve the traditional sparse representation classifier. It can effectively solve the nonlinear problems and make the test samples accurately represented as a sparse linear combination of training samples. Finally, a new algorithm is proposed to improve the performance of kernel sparse representation classification algorithm. This new algorithm called a weighted kernel sparse representation classification algorithm based on the local constraints can improve the classification accuracy. Then experiments prove the correctness of this idea.(5) The improvement of the kernel function of support vector machine is proposed. In order to highlight the different effects on the classification of different features, this paper first adds the weights of feature important information to the polynomial kernel function and the Gauss kernel function. Then a new combined kernel consisted of the improved polynomial kernel function and the Gauss kernel function is established. At last, this paper obtains optimal kernel parameters through the parameters optimization method. This algorithm improves the basis function, uses the combined kernel function to replace single kernel function and finds the optimal parameters of kernel and combination of parameters.It makes multiple changes to improve the performance of the algorithm. Finally, experiments demonstrate the good performance of the improved algorithm.