摘要
为了提高语音情感的正确识别率,在情感语音韵律特征的基础上,提出情感语音音质特征的提取。结合音质特征参数和韵律特征参数,采用支持向量机分类器实现汉语普通话生气、高兴、悲伤和惊奇四种主要情感类型语音的情感识别。实验结果表明,语音音质特征参数和韵律特征参数相结合取得的情感平均正确识别率为88.1%,比单独使用韵律特征参数高出6%。可见,语音音质特征是一种较有效的情感特征参数。
Based on extracting fundamental prosody features from emotional speech, speech voice quality features are extracted to improve emotion recognition accuracies. Utilizing support vector machines, we recognize main four speech emotions like anger, happiness, sadness and surprise in Chinese mandarin emotional speech corpus by combining voice quality and prosody features. The experimental results show that, the single prosody features yield an overall accuracy of 86.1%, whereas combining voice quality and prosody features yields an overall accuracy of 88.1%, making an approximately 6% improvement for emotion recognition. It also shows that speech voice quality features are effective emotional features.
出处
《电路与系统学报》
CSCD
北大核心
2009年第4期120-123,共4页
Journal of Circuits and Systems
基金
浙江省教育厅高校青年教师资助(ZX2005)
关键词
韵律特征
音质特征
支持向量机
语音情感识别
prosody features
voice quality features
saapport vector machines
speech emotion recognition
