基于声门特征与语音特征的语音情感识别对比研究  

A Comparison Study on Emotion Recognition with Glottal Features and Speech Features

作　　者：朱智豪 郑徵羽 肖仲喆 ZHU Zhihao;ZHENG Zhiyu;XIAO Zhongzhe(School of Optoelectronic Science and Engineering,Soochow University,Suzhou 215006,China)

机构地区：[1]苏州大学光电科学与工程学院,江苏苏州215006

出　　处：《电声技术》2020年第5期41-46,共6页Audio Engineering

基　　金：国家自然科学基金青年基金项目(61906128,61802272);江苏省自然科学基金青年基金项目(BK20180834)。

摘　　要：语音发声过程的物理过程对于理解语音情感的表达机制具有重要意义。本文研究了声门波信号作为语音激励源在语音情感表达中的贡献。本文采用对比研究的方式,从完整语音信号与声门波中提取相同定义的特征,并使用相同的机器学习算法进行情感识别实验。实验结果表明,声门波信号能够携带语音信号中的大部分情感信息,尤其在唤醒度方向对语音情感能够较充分地表达;而声门波在语音情感的价效度方向有明显的信息缺失,需要依赖声道特性对情感价效进行表达。The physical process of speech production plays an important role in understanding of the expression mechanism of vocal emotion.In this paper,we studied the contribution of glottal waveform as the source of speech stimulation to expression of speech emotion.The same feature set is extracted from the speech signal and glottal waveform for automatic emotion recognition with the same machine learning algorithm.The experimental results prove that glottal waveform can carry most of the emotional information in speech signals,especially in the arousal dimension.While glottal waveform has obvious information deficiency in the valence dimension,which relies more on vocal tract property.

关 键 词：声门波信号 语音信号 语音情感识别 

分 类 号：TN912.3[电子电信—通信与信息系统]