基于基频特征的情感语音识别研究  

Research on emotional speech recognition based on pitch

作　　者：郭鹏娟[1] 蒋冬梅[1] Hichem Sahli Werner Verhelst 

机构地区：[1]西北工业大学计算机学院,西安710072 [2]比利时布鲁塞尔自由大学电子与信息工程系

出　　处：《计算机应用研究》2007年第10期101-103,共3页Application Research of Computers

基　　金：西北工业大学基金资助项目(04XD0102);中国科技部与比利时弗拉芒大区科技合作资助项目(国科外函[2004]487)

摘　　要：探索在不同的情感状态下的基音特征变化规律。通过对含有生气、高兴、悲伤情感语音信号进行分析,总结了情感语音基频的变化规律,确定了用于情感识别的12维的基频的基本特征以及扩展特征,运用混合高斯模型进行情感识别,并作了识别实验,获得了较好的结果。This paper first analyzed the change of pitch in speech signal under four emotions such as angry, happy, neutral and sad. Then twelve pitch related emotional features were extracted and used to train the Gaussian mixture models （GMMs） for each emotional state. In emotion recognition, the emotional state with the GMM having the maximal likelihood ratio was determined as the final out. The experimental result shows that the pitch related features and GMM are effective in emotion recognition from speech.

关 键 词：语音信号 基频 情感特征 情感识别 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]