摘要
目的研究语音特征梅尔频率倒谱系数(MFCC)的选取对说话人识别系统性能的影响。方法采用基于平均影响值(MIV)的支持向量机(SVM)方法研究了说话人识别中的梅尔频率倒谱系数各维倒谱分量对于识别分类的贡献度。结果选择具有代表性的特征向量进行说话人分类识别,能得到维数更低、识别率更高的特征参数。结论通过MIV值可判断各维特征参数分量的重要性,选取权重值高的MFCC特征参数来提高系统识别率和缩短系统运行时间。
Aim To study the influence of the selection of MFCC on the performance of speaker recognition system. Methods The method of SVM based on the MIV is used to study the contribution of each dimension cepstrum component of MFCC in speaker recognition to the identification and classification. Results The selection of the representative feature vector for speaker classification and recognition can result characteristic parameters with lower dimensions and higher recognition rate. Conclusion The importance of each dimension feature parameters＇s com- ponent can be judged by MIV. Selecting MFCC with high weight value can improve the system recognition and reduce system operation time.
出处
《西北大学学报（自然科学版）》
CAS
CSCD
北大核心
2013年第2期203-208,共6页
Journal of Northwest University（Natural Science Edition）
基金
国家自然科学基金资助项目(11074159
11074158)
陕西师范大学研究生培养基金资助项目(2012CXS034)
关键词
说话人识别
语音特征参数
梅尔频率倒谱系数
支持向量机
平均影响值
speaker recognition
voice feature parameter
reel-frequency cepstral coefficient
support vector machine
mean impact value
