摘要
根据语音发声过程中的混沌特性,应用非线性动力学模型分析情感语音信号,提取了该模型下情感语音信号的非线性特征以及常用的声学特征(韵律特征和MFCC).设计情感语音识别对比实验,将非线性特征与不同声学特征融合并验证了该组合下的情感识别性能,研究了语音信号混沌特性对情感语音识别性能的影响.实验选用德国柏林语音库4种情感(高兴、愤怒、悲伤和中性)作为语料来源,支持向量机网络用于情感识别.结果表明,非线性特征有效表征了情感语音信号的混沌特性,与传统声学特征结合后,情感语音识别性能得到了显著提高.
Based on the chaotic characteristics of emotional speech,nonlinear features and frequently used acoustic features were extracted to effectively differentiate emotions by applying a nonlinear dynamic model to analyzethe emotional speech signals.The effectiveness of nonlinear features was verified by comparison with the integrated model of nonlinear features with different acoustic features（prosodic features and MFCC）on the recognition rates of emotional speech.It also studied the influences of chaotic characteristics of speech signals on the recognition rates of emotional speech.Four types of emotion（happiness,anger,sadness,and neutrality）from Berlin databasewere selected and support vector machine was used for emotion recognition.The results show the nonlinear features effec-tively represent the chaotic characteristics of emotional speech signals.The recognition rates of emotional speech can be significantly improved when nonlinear features are combined with traditional acoustic features.
出处
《天津大学学报（自然科学与工程技术版）》
EI
CAS
CSCD
北大核心
2015年第8期681-685,共5页
Journal of Tianjin University：Science and Technology
基金
国家自然科学基金资助项目(61371193)
山西省青年科技研究基金资助项目(2013021016-2)
山西省回国留学人员科研资助项目(2013-034)
关键词
情感语音识别
混沌特性
支持向量机
非线性特征
emotional speech recognition
chaotic characteristic
support vector machine
nonlinear feature
