摘要
针对MFCC滤波器存在语音高频信号泄露的问题,为避免基于MFCC特征对语音进行情感识别时存在有效情感特征丢失的局限性,结合MFCC的高准确性和GFCC的强鲁棒性,提出了基于MFCC与GFCC混合特征训练CNN对语音进行情感识别的方法,有效提高了语音情感识别的准确率,改善了CNN模型的识别性能。实验结果表明,所设计的混合特征识别方法较传统识别方法识别率明显升高并达到了83%,实现了语言情感识别准确率的有效提升。
Aiming at the problem of voice high frequency signal leakage in Mel-scale frequency cepstral coefficients(MFCC)filter,in order to avoid the limitation of effective emotional feature loss when emotion recognition based on MFCC feature,combined with the high accuracy of MFCC and the strong robustness of GFCC,based on the hybrid feature of MFCC and GFCC,CNN is used to identify the emotion of speech,which improves the accuracy of speech emotion recognition and improves the recognition performance of CNN model.Experimental results show that the proposed hybrid feature recognition method has a significantly higher recognition rate than the traditional recognition method and reaches 83%,which achieves an effective improvement of the language emotion recognition accuracy.
作者
郭卉
姜囡
任杰
GUO Hui;JIANG Nan;REN Jie(Criminal Investigation Police University of China,Shenyang 110854,China)
出处
《光电技术应用》
2019年第6期34-39,共6页
Electro-Optic Technology Application
关键词
MFCC
GFCC
语音情感识别
CNN
混合特征
Mel-scale frequency cepstral coefficients(MFCC)
Gammatone frequency cepstral coefficients(GFCC)
speech emotion recognition
cellular neural network(CNN)
mixed features
作者简介
郭卉(1996-),女,湖北武汉人,硕士研究生,主要研究方向为视听资料检验技术;姜囡(1979-),女,山东武城人,博士,副教授,硕士研究生导师,主要研究方向为公安视听技术及模式识别;任杰(1995-),男,陕西黄陵人,硕士研究生,主要研究方向为视听资料检验技术。
