混合MFCC特征参数应用于语音情感识别  

Speech Emotion Recognition Based on Mixed MFCC Characteristic Parameter

作　　者：周萍[1] 李晓盼[1] 李杰[2] 景新幸[3] 

机构地区：[1]桂林电子科技大学电子工程与自动化学院,广西桂林541004 [2]桂林电子科技大学计算机科学与工程学院,广西桂林541004 [3]桂林电子科技大学信息与通信学院,广西桂林541004

出　　处：《计算机测量与控制》2013年第7期1966-1968,1986,共4页Computer Measurement &Control

基　　金：国家自然科学基金资助项目(60961002);广西自然科学基金资助项目(2012GXNSFAA053221)

摘　　要：引入两种新的特征参数Mid-MFCC和IMFCC,采用MFCC、Mid-MFCC和IMFCC相结合的改进算法,解决MFCC特征参数在语音识别中对中、高频信号的识别精度不高的特点,并使用增减分量法计算MFCC、Mid-MFCC和IMFCC各阶倒谱分量对语音情感识别的贡献,提取3个特征参数贡献最高的几阶倒谱分量组成了新的特征参数;实验结果表明,在相同环境下新的特征参数比经典MFCC特征参数的语音情感的识别率稍高。This paper introduced two new characteristic parameters Mid-- MFCC and IMFCC combining with MFCC to improve the algo- rithm which solve the problem that MFCC characteristic parameter in speech recognition has low identification accuracy when signal is inter- mediate, high frequency signal, calculating the contribution that MFCC, Mid--MFCC and !MFCC each order cepstrum component was used in speech emotion recognition with increase or decrease component method, extracting highest contribution of several number order cepstrum component from three characteristic parameters and forming a new characteristic parameter. The experiment results show that new character- istic parameter has higher recognition rate than classic MFCC characteristic parameter in speech emotion recognition under the same environ- ment.

关 键 词：Mel频率倒谱系数(MFCC) 增减分量法 特征提取 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]