摘要
为了有效提高语音情感识别的准确率,结合Mel频率倒谱系数特征准确性高、GFCC特征抗噪性强的特点,提出了一种将GFCC与Mel频率倒谱系数、韵律特征相融合的语音情感识别算法。针对高兴、悲伤、惊恐、中性、生气5种情感语音,分别对单一Mel频率倒谱系数特征算法、单一GFCC特征算法及改进的混合特征算法进行实验比对分析。实验结果表明,加入GFCC的混合特征后,语音情感识别算法的识别准确率及稳定性均有明显的提高,对公安工作具有现实的意义。
In order to effectively improve the accuracy of speech emotion recognition,combining the characteristics of the Mel-frequency Cepstral Coefficients(MFCC)with high accuracy and theγ-Tone Filter Cepstral Coefficients(GFCC)with strong anti-noise ability,a speech emotion recognition Algorithm based on GFCC,MFCC and prosody features is proposed to improve the accuracy of speech emotion recognition in complex backgrounds.Experimental results show that the recognition accuracy and stability of the speech emotion recognition algorithm are obviously improved after adding GFCC’s mixed features,which is of great practical significance to the police work.
作者
王华朋
刘恩
晁亚东
刘元周
倪令格
WANG Huapeng;LIU En;CHAO Ya-dong;LIU Yuan-zhou;NI Ling-ge(Department of Audio-Visual Data Inspection Technology,Criminal Investigation Police University of China,Liaoning Shenyang 110035)
出处
《中国刑警学院学报》
2020年第2期124-128,共5页
Journal of Criminal Investigation Police University of China
基金
公安部公安理论及软科学(编号:2017LLYJXJXY040)
重庆市高校刑事科学技术重点实验室(西南政法大学)开放基金(编号:XKZDSYS2019-Z1)
上海市现场物证重点实验室开放课题基金(编号:2018XCWZK09)。
关键词
GFCC
SVM
语音情感特征
情感识别
γ-Tone Filter Cepstral Coefficients(GFCC)
Support Vector Machine(SVM)
Speech emotion feature
Emotion recognition
