Use of spectral subband moments in MFCC computation

Author

Gjelsvik, Eigil ; Paliwal, Kuldip K.

Author_Institution

Sch. of Microelectron. Eng., Griffith Univ., Brisbane, Qld., Australia

Volume

2

fYear

1999

fDate

1999

Firstpage

637

Abstract

Mel frequency cepstral coefficients (MFCCs) are currently the most popular form of parameterization of the speech signal in speech recognition systems. In this paper, we look at a way to improve the extraction of these features using information about the spectral characteristics of the signal to modify filter-bank shapes. This information is captured in the form of spectral moments of the subbands. We show that this improves speech recognition performance, but the improvement is not very significant

Keywords

cepstral analysis; channel bank filters; feature extraction; filtering theory; speech recognition; MFCC computation; feature extraction; filter-bank shapes; mel frequency cepstral coefficients; spectral characteristics; spectral subband moments; speech recognition performance; speech recognition systems; speech signal parameterization; Additive white noise; Australia; Cepstral analysis; Data mining; Filters; Frequency estimation; Mel frequency cepstral coefficient; Shape; Signal processing; Speech recognition;

fLanguage

English

Publisher

ieee

Conference_Titel

Signal Processing and Its Applications, 1999. ISSPA '99. Proceedings of the Fifth International Symposium on

Conference_Location

Brisbane, Qld.

Print_ISBN

1-86435-451-8

Type

conf

DOI

10.1109/ISSPA.1999.815753

Filename

815753