Improved MFCC Algorithm in Speaker Recognition System

摘要：

In speaker recognition systems, one of the key feature parameters is MFCC, which can be used for speaker recognition. So, how to extract MFCC parameter in speech signals more exactly and efficiently, decides the performance of the system. Theoretically, MFCC parameters are used to describe the spectrum envelope of the vocal tract characteristics and often ignore the impacts of fundamental frequency. But in practice, MFCC can be influenced by fundamental frequency which can cause palpable performance reduction. So, smoothing MFCC (SMFCC), which based on smoothing short-term spectral amplitude envelope, has been proposed to improve MFCC algorithm. Experimental results show that improved MFCC parameters—SMFCC can degrade the bad influences of fundamental frequency effectively and upgrade the performances of speaker recognition system. Especially for female speakers, who have higher fundamental frequency, the recognition rate improves more significantly.

关键词： speaker recognition MFCC spectral envelop five-spot triple smoothing

作者: Yibo SHI Li WANG

作者单位: School of Computer & Science Guizhou University Guiyang, China

会议类型: 国际会议

会议名称: 2010 International Conference on Signal and Information Processing(2010年IEEE信号与信息处理国际会议 ICSIP2010)

会议地点: 长沙

会议语种:英文

页码: 330-333

在线出版日期: 2010-12-14（万方平台首次上网日期，不代表论文的发表时间）