采用半定规划多核SVM的语音情感识别

Speech Emotion Recognition Using Semi-Definite Programming Multiple-Kernel SVM

为提高语音情感识别精度,采用二叉树结构设计多分类器,其中使用半定规划法求解并构造多核支持向量机(SVM)分类模型,并采用均方根误差与最大误差对分类器性能进行衡量.对特征选择之后的参数集合进行了测试,结果表明,采用半定规划多核SVM分类模型的情感识别精度达到88.614%,比单核分类模型的识别精度提高了12.376%,且能有效减少误差积累和降低情感状态之间混淆程度.

To improve the accuracy of speech emotion recognition, a multi-class classifier with binary- tree structure is adopted, which includes building the multi-kernel support vector machine （SVM） classi- fier model solved by semi-definite programming method, and using the root mean square error and maxi- mum error to evaluate the performance of the classifier. Through the test on the parameter set obtained by feature selection algorithm, the results of experiments show that the total recognition accuracy of the pro- posed multiple-kernel SVM classifier model using semi-definite programming is 88. 614%, which is 12. 376% higher than that of single-kernel SVM model. Moreover the multiple-kernel SVM model can re- duce the total error accumulation and confusion between emotion states.