Speech Emotion Recognition based on Optimized Support Vector Machine

摘要

Speech emotion recognition is a very importantspeech technology. In this paper, Mel Frequency CepstralCoefficients (MFCC) has been used to represent speechsignal as emotional features. MFCCs plus energy of anutterance are used as the input for Support Vector Machine.Support Vector Machine (SVM) has been profoundlysuccessful in the area of pattern recognition. In the recentyears there has been use of SVM for speech recognition.Many kinds of kernel functions are available for SVM tomap an input space problem to high dimensional spaces. Welack guidelines on choosing a better kernel with optimizedparameters of SVM. Some kernels are better for somequestions, but worse for other questions. Which is better isunknown for speech emotion recognition, thus the thesisstudies the SVM classifier and proposes methods used toselect a better kernel with optimized parameters. The newmethod we proposed in this paper can more efficiently gainoptimized parameters than common methods. In order toimprove recognition accuracy rate of the speech emotionrecognition system, a speech emotion recognition based onoptimized support vector machine is proposed.Experimental studies are performed over the HITEmotional Speech Database established by SpeechProcessing Lab in School of Computer Science andTechnology at HIT. The experiment result shows that thespeech emotion recognition based on optimized SVM canimprove the performance of the emotion recognition systemeffectively