【作者】 李向伟； 方强； 李爱军； 王红；

【Author】 Li Xiangwei 1,Fang Qiang 2,Li Aijun 2,Wang Hong 1 1.College of Information Science and Technology,Shandong Normal University,Jinan 250014,China 2.Institute of Linguistics,Chinese Academy of Social Sciences,Beijing 100732,China

【机构】 山东师范大学信息科学与工程学院； 中国社会科学院语言研究所；

【摘要】 本文主要寻找嗓音音质中能够区分情感的因素,为下一步的情感语音合成作准备。我们基于同一发音人的七种不同情感(七种情感分别为:悲伤、高兴、害怕、厌恶、生气、惊讶、中性)语音样本提取了基频抖动jitter、振幅抖动shimmer、谐波噪声率HNR、基频均值meanF0、声门波震动幅度PulseAmp、声门波形最大下降率MFDR等与嗓音声源密切相关的8个声学参数并进行统计分析。结果表明在不同情感下一些参数如NAQ,MFDR具有显著性差异,而其他参数如shimmer,h1-h2差异较小。在两种具体情感对组合的分析过程中,各个参数表现出的差异性也有所不同。更多还原

【Abstract】 In this study,we mainly looked for the distinguishable factors in voice quality,to prepare for emotional speech synthesis.Eight parameters which are based on one speaker’s voice samples in seven different kinds of emotions(They are sad,joy,fear,disgust,surprise,anger and neutral) were extracted and analyzed.They are jitter,shimmer,HNR,mean F0,Pulse Amp,MFDR,NAQ,h1-h2,which have close relationship with glottal voice.A series of statistic analysis showed that some of these parameters such as NAQ and MFDR are more significant in distinguishing emotions and in pair-wise comparisons between each emotion pair than others.更多还原