基于MFCC0的语音端点检测方法  

Endpoint Detect Method Based on MFCC0

作　　者：舒倩[1] 李银国[1] 

机构地区：[1]重庆邮电大学自动化学院,重庆400065

出　　处：《通信技术》2007年第11期374-375,378,共3页Communications Technology

基　　金：国家自然科学基金资助项目(编号:69803014);国家"863"计划(编号:2004AA1Z2380)

摘　　要：MFCC是语音识别中常用的特征参数,根据MFCC分量对语音端点的敏感性,提出利用平常舍去的识别特征参数分量MFCC0作为语音端点检测的参量。接着根据MFCC0的特性设计了一种新的端点检测方法,该方法简单且无需增加额外的计算量。实验结果表明,基于该方法的语音识别系统不仅可以通过端点检测大大压缩数据量,而且提高了系统的识别率。MFCC is always used as feature parameter in speech recognition system. The paper proposes a new method based onMFCC0, which is eliminated as parameter for speech recognition, according to sensitivity analysis on components of MFCC. The new method is simple, and the additional calculation is very little. The results of experiments show that the speech recognition system can greatly reduce calculation amount and achieve higher recognition rate by using endpoint detect method based on MFCC0.

关 键 词：语音识别 语音端点检测 MFCC CHMM 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]