摘要
目前对MFCC的应用与研究,一般忽略了基音频率对MFCC的影响.分析发现,基音频率会影响MFCC对声道特性的准确描述,进而影响说话人识别系统的性能;由此提出了一种基于平滑幅度谱的SMFCC(smoothing MFCC),在YOHO说话人识别数据库上的实验表明,SMFCC性能在整体上优于MFCC,而在女性说话人数据集上性能提高尤其明显,并且具有更好的时间鲁棒性.
Fundamental frequency can affect the accurate description of MFCC and degrade the performance of speaker recognition system, so a new parameter called SMFCC （smoothing MFCC） was proposed. The parameter mainly employs the method of smoothing amplitude spectral to reduce the effect of fundamental frequency. Experiments in YOHO speech database show that SMFCC is more effective and robust than the conventional MFCC, especially in the female speaker recognition system.
作者简介
陆伟，男，1969年生，博士生／讲师．研究方向：语音信号处理．E-mail：luwei@ustc．edu．cn
通讯作者：戴蓓荷，教授．E-mail：bqdai@ustc．edu．cn
