Mandarin Speech Emotion Recognition Based On HMM And Artificial Neural Network Hybrid Model

With the development of human-computer interaction technology, speech emotion recognition has gradually become a hot topic in the field of pattern recognition. With the deepening of the research, the recognition effect was not very ideal when just used HMM or Artificial Neural Network to classify speech emotional status. In this paper, HMM and Artificial Neural Network algorithms were combined with each other to overcome its shortcoming and improve the recognition rate of the system. This recognition technology could widely be used in robots system and automated voice response system.In this paper, the development of speech emotion recognition was analyzed, and mandarin speech emotion recognition problems were studied. First of all, because it is very difficult to establish mandarin emotional speech database, the basic theory of speech emotion and the principle of establishing database were detailed in this paper. Then, in order to reduce the effects of the speech signal itself, the method based on GA and SVM was proposed to detect the endpoint of speech, and the emotional features of speech were analyzed at the same time. The feature vector with 30 parameters was extracted in this paper. The HMM and artificial neural network hybrid model was put forward to make up for the defects of HMM and artificial neural network model, and apply it to the research of mandarin speech recognition of emotional state. It firstly got the best sequence of emotional speech signal by HMM. Then, the feature parameters of the same state were structured as uniform dimension by Time Warping algorithm, and use it as the input of Artificial Neural Network to recognition the speech emotional states. The strong ability of HMM to establish the model for dynamic time sequence and the powerful classification ability of Artificial Neural Network were fully used in the hybrid model.On the Matlab2012 a platform, the experiments were simulated. The combination of HMM and artificial neural network model was realized, and the hybrid model were compared with HMM and Artificial Neural Network. The results show that the HMM and Artificial Neural Network hybrid model could improve the recognition rate of mandarin speech emotion recognition system. In addition, its advantages were reflected in different SNR.