基于生理声学的语音情感评价尺度研究  

A Study on Speech Emotion Evaluation Scale Based on Physiology and Acoustics Features

作　　者：杨洁[1] 梁昌维 吴西愉 孔江平[1] YANG Jie;LIANG Changwei;WU Xiyu;KONG Jiangping

机构地区：[1]北京大学语言学实验室/北京大学中文系,北京100871

出　　处：《语言学论丛》2023年第4期3-19,共17页Essays on Linguistics

基　　金：国家社会科学基金重大项目“中华民族语言文字接触交融研究”(22&ZD213)的支持。

摘　　要：语音的生理声学特征是感知语音情感变化的重要线索。近年来,关于语音情感的声学特征研究主要集中在韵律特征、语谱特征以及音质特征上,随着研究的深入,出现了部分嗓音特征相关研究。本文首先基于语音信号和嗓音生理信号,进行了基频、振幅、开商、速度商、第一二三共振峰、语速和音速等参数的提取,构建了爱、憎、喜、悲、急、惧、怒、中八种语音情感的快速情感分类模型。其次,将上述参数分为韵律参数、共鸣参数和嗓音参数,分析了语音情感评价尺度模型中各组参数对于模型评估能力的影响。研究结果表明,联合使用韵律参数、共鸣参数和嗓音参数比单独使用每类参数或每两类参数具有更高的预测精度。对于单独--种语音情感的评估来说,并不是参数越多评估效果越好。因此,在语言学上语音情感的感知范畴体现为复杂的多区别性特征系统。The acoustic features of speech are important clues to perceive emotional changes.In recent years,studies on the acoustic features of emotional speech have mainly focus on prosodic features,spectral based features and quality features.More in-depth research have investigated specific voice characteristics.Based on speech signals and voice signals,this paper extracted fundamental frequency,amplitude,open quotient,speed quotient,the first three formants,speaking rate and articulation rate,and constructed a fast classification model for eight kinds of emotional speech,namely,love,hate,joy,sadness,fear,urgency,anger and normal.The above parameters are divided into prosody parameters,resonance parameters and voice parameters,and then the influence of each group of parameters on the evaluation ability in the speech emotion evaluation scale is analyzed.We found that the combined use of prosody parameters,voice parameters and resonance parameters had higher prediction accuracy than the model constructed by using one single type of parameters.For the evaluation of a single speech emotion,it is not that the more parameters,the better the evaluation effect.Therefore,in linguistic terms the categorical perception of speech emotion is reflected as a complex system of multiple distinctive features.

关 键 词：语音情感评价尺度 生理声学特征 韵律参数 嗓音参数 共鸣参数 

分 类 号：H11[语言文字—汉语]