改进MFCC参数在非特定人语音识别中的研究  

Research of Improved MFCC Parameters in Signer-independent Speech Recognition

作　　者：田莎莎[1] 唐菀[1] 佘纬[2] 

机构地区：[1]中南民族大学计算机科学学院,武汉430074 [2]中南民族大学数学与统计学院,武汉430074

出　　处：《科技通报》2013年第3期139-142,146,共5页Bulletin of Science and Technology

基　　金：国家自然科学基金资助项目(61103248);中南民族大学自然科学基金资助项目(YZQ10004)

摘　　要：随着信息时代的高速发展,人们越来越关注计算机的便携使用方式,以语音输入代替手动输入成为计算机未来发展的一个必然趋势。本文在MFCC特征参数的基础上,提出了一种改进MFCC特征参数--BMFCC特征参数,以提高原MFCC特征参数在语音识别时的识别率和运算速度。BMFCC特征参数在进行参数的提取时,分为特征分量加权、特征分量求差分、主成分分析三个步骤。仿真实验结果表明,本文提出的BMFCC特征参数在识别率和有运算速度上均优于MFCC特征参数,且更具鲁棒性。With the rapid development of the information age,people pay more and more attention to the portable use of computer,taking voice input instead of manual input will become an inexorable trend of computer development.The paper proposed an improved MFCC feature parameters-BMFCC characteristic parameters on the basis of MFCC feature parameters to improve the recognition rate and operation speed of original MFCC characteristic parameters in speech recognition.In the extraction of BMFCC characteristic parameters,it was divided into three steps of feature component weight,characteristic component difference algorithm and principal component analysis.The simulation results show that the proposed BMFCC characteristic parameter is superior and more robust to the MFCC characteristic parameters in the recognition rate and operation speed.

关 键 词：MFCC参数 BMFCC参数 语音识别 非特定人 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]