基于多分类器集成的语音情感识别

Speech Emotion Recognition Based on Integration of Multiple Classifiers

为了提高语音情感的正确识别率,提出一种基于多分类器集成的语音情感识别新算法.首先提取情感语音的韵律特征、音质特征和MFCC特征参数,然后将贝叶斯网络、K近邻法和径向基神经网络三种分类器构成集成分类器,实现对Berlin情感语言数据库中愤怒、欢乐、悲伤、中性、恐惧、无聊和厌恶7种主要情感类型的识别.实验结果表明,集成分类器对语音情感的识别取得了71.4019%的平均正确识别率,识别效果优于单一分类器.

A new method of speech emotion recognition via integration of multiple classifiers is proposed for improving speech emotion classification rate. Based on extracting prosody, voice quality and MFCC feature parameters from emotional speech, three kinds of classifiers including Bayes. net, K-Nearest Neighborhood （KNN） and Radial Basis Function （RBF） neural network are utilized to construct an ensemble classifier so as to realize recognizing the seven main speech emotion in Berlin emotional speech database like anger, i oy, sadness and neutral, fear, bore and disgust. Computer simulation results show that the ensemble classifier can achieve average correct rate of 71. 4019 % for speech emotion classification, which is superior to the single classifier.