Research On Speech And Emotional Recognition

With the rapid development of artificial intelligence and information technology, people have become increasingly eager that computers should possess the capability of intelligentized man-machine interaction, and speech is the most common, effective and convenient way for communication. In addition to expressing the basic semantic information, people also phonate through vocal cords to convey their emotions, moods and so forth. Effective methods for recognizing the emotional information existing in the speech are of great significance to enhance intelligentized and humanized levels of computers. Currently, the speech emotion recognition technology has been extensively applied to the field of education, information, medical science, criminal investigation, entertainment and so on.The major work of this paper is as follows:(1) Use minimum variance distortionless response (MVDR) to extract speech emotion feature parameters. In view of the problems that liner prediction (LP) model cannot well express the envelope of the speech emotion spectrum, which will result in low recognition rate, the MVDR spectrum method was introduced in this paper for extraction of speech emotion feature parameters. Firstly, Levinson-Durbin algorithm was used to calculate the M-order linear prediction coefficient. Then, calculated the MVDR spectrum coefficient, which would be filtered subsequently through a Mel scaled triangular filter to obtain the logarithmic energy output from each filter. Finally, MVDR feature was obtained through discrete cosine transform.(2) Use artificial bee colony algorithm (ABC) to select speech emotion feature parameters. In view of the shortcomings of excessive redundant information and system uptime when using minimum variance distortionless response algorithm for speech emotion feature extraction, the artificial bee colony algorithm (ABC) was used to select speech emotion feature parameters. And we selected the optimal speech emotion feature subset as the characteristic parameters for recognition.(3) The pulse-coupled neural network (PCNN) and radial basis function (RBF) were combined for classification and recognition of speech emotion. In view of the shortcomings that the single classifier PCNN and RBF neural network will acquire low speech emotion recognition rate, the method of combination of PCNN and RBF for speech emotion recognition was proposed in this paper. Firstly, spectrogram algorithm was applied to obtain the spectrogram of speech emotion. Then, the spectrogram was fed to PCNN to get the characteristic time sequences, which were regarded as the characteristic parameters used for speech emotion recognition. Eventually, RBF was adopted for speech emotion recognition.For the methods used above, the training and testing samples were both taken from the Chinese CASIA emotional corpus which contains four types of speech emotions: angry, calm, happy and fear. Then, MATLAB was used for simulation experiment. As the experimental result indicated, the proposed method was effective to improve the recognition rate of speech emotion.