基于多分类器投票组合的语音情感识别

为了提高语音情感的正确识别率,提出一种基于多分类器投票组合的语音情感识别新方法.在提取情感语音的韵律特征和音质特征基础上,利用投票方法将支持向量机、K近邻法和人工神经网络三种分类器构成组合分类器,实现对汉语生气、高兴、悲伤和惊奇4种主要情感类型的识别.实验结果表明,与使用单一分类器相比,组合分类器对语音情感的识别取得了87.4%的平均正确识别率,识别效果优于单一分类器.

Speech Emotion Recognition Based on Voting Combination of Multiple Classifiers

A new method of speech emotion recognition via voting combination of multiple classifiers is proposed for improving speech emotion classification rate. Based on extracting prosody and voice quality features from emotional speech, three kinds of classifiers including Support Vector Machines(SVM), K-Nearest Neighborhood(KNN) and Artificial Neural Network(ANN) are utilized to vote and construct an ensemble classifier so as to realize recognizing the four main speech emotion in Chinese like anger, happiness, sadness and surprise. Compared with the single classifier, computer simulation results show that the ensemble classifier can achieve average correct rate of 87.4% for speech emotion classification, which is superior to the single classifier.