Frequency-Warping Invariant Features for Automatic Speech Recognition

Author

Mertins, Alfred ; Rademacher, Jan

Author_Institution

Dept. of Phys., Oldenburg Univ.

Volume

5

fYear

2006

fDate

14-19 May 2006

Abstract

Based on the well-known relationship between vocal tract length (VTL) variation and linear frequency warping, we present a method for generating vocal tract length invariant (VTLI) features. These features are computed as translation invariant, correlation-type features in a log-frequency domain. In phoneme classification and recognition experiments on the TIMIT database, their discrimination capabilities and robustness to mismatches between training and test conditions turned out to be considerably better than for Mel-frequency cepstral coefficients (MFCCs). The best results are obtained when VTLI features and MFCCs are combined

Keywords

cepstral analysis; correlation methods; frequency-domain analysis; signal classification; speech recognition; wavelet transforms; Mel-frequency cepstral coefficients; TIMIT database; automatic speech recognition; correlation-type features; discrimination capabilities; frequency-warping invariant features; linear frequency warping; log-frequency domain; phoneme classification; recognition experiments; translation invariant; vocal tract length variation; Automatic speech recognition; Bandwidth; Continuous wavelet transforms; Discrete wavelet transforms; Fourier transforms; Frequency; Hidden Markov models; Robustness; Testing; Wavelet transforms;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech and Signal Processing, 2006. ICASSP 2006 Proceedings. 2006 IEEE International Conference on

Conference_Location

Toulouse

ISSN

1520-6149

Print_ISBN

1-4244-0469-X

Type

conf

DOI

10.1109/ICASSP.2006.1661453

Filename

1661453