Modified MFCC Methods Based on Differential Power Spectrum and Power Law for Robust Speech Recognition

被引：0

作者：

Li, Gao-yun ^{[1
]}

Wang, Xiong ^{[2
]}

机构：

[1] Northwest Univ Nationalities, Sch Elect Engn, Lanzhou, Gansu, Peoples R China

[2] Northwest Normal Univ, Coll Educ Technol, Lanzhou, Gansu, Peoples R China

来源：

COMPUTER SCIENCE AND TECHNOLOGY (CST2016) | 2017年

关键词：

Differential power spectrum; Feature extraction; MFCC; Speech recognition;

D O I：

暂无

中图分类号：

TP301 [理论、方法];

学科分类号：

081202 ;

摘要：

This paper proposes a new feature for robust speech recognition. This method combines the benefits of differential power spectrum and the power law. The extracted features from this combined method are known as mel frequency differential power cepstral coefficients (MFDPCC).The speech recognition performance of MFDPCC features is compared to the conventional mel frequency cepstrum coefficients (MFCC) features using fuzzy radial basis function (FRBF) neural network under different noise level and different vocabulary. The experiment results demonstrate that the proposed MFDPCC feature has higher recognition rate and better robustness than MFCC.

引用

页码：696 / 702

页数：7