The Research On Speech Emotion Recognition Based On ITD

In order to make the people communicate with computer more humanized and natural,the computer having the function of emotion recognition gradually becomes the focus of research in various fields. Among them, speech emotion recognition plays an important role in emotion recognition. In order to improve the accuracy and efficiency of speech emotion recognition, this paper improves the characterization of speech emotion features deeply, and uses new way to improve the identification method. The specific research work is as follows:(1)For the speech signal keeping linear smooth in short-time, we extract the traditional characteristics of short-time energy,average zero crossing rate, pitch frequency, average magnitude,the former 12 order MFCC coefficient from the speech signal of four basic feelings(Angry, Happiness, Neutral, Sadness).Through computing statistical characteristics of these characteristics,we get a total of 92 dimensions of traditional characteristics.(2)The speech is changeable and nonlinear in nature, containing rich emotional information. To describe the speech signal more comprehensive by extracting the inconstant and nonlinear features, we make the speech signal do Intrinsic Time scale of Decomposition(ITD) to get several former order proper rotation(PR) component, and then extract the instantaneous parameters of the first three order PR component and correlation dimension of PR1 component, which are as the new emotional characteristics of speech signal.(3)In recognition experiments, this paper uses the support vector machine(SVM) to identify four groups of combination features. By comparing the recognition rate of each experiment, we find that the recognition rate of group with the instantaneous characteristic and the correlation dimension features has been effectively improved. But the similar emotions still exist mutual miscalculation. For the additional momentum Back Propagation Neural Network(BP neural network- BPNN) doing identification having quick calculation speed, strong reliability and good stability characteristics, we apply it to the forth experiment as well. The experimental results show that the new algorithm has better emotion recognition rate and can effectively reduce the similar emotional misjudgment.