Home // International Journal On Advances in Systems and Measurements, volume 2, number 1, 2009 // View article

Performance of Spectral Amplitude Warp based WDFTC in a Noisy Phoneme and Word Recognition Tasks

Authors:
Rangarao Muralishankar
H. N. Shankar

Keywords: Robustness, Speech recognition, Warped Discrete Fourier Transform, Cepstrum, WDFTC, PMVDR, Spectral AmplitudeWarping, WDFTC SAW

Abstract:
In this paper, we investigate the noise robustness of three features, namely, the warped discrete Fourier transform cepstrum (WDFTC, [1]), perceptual minimum variance distortionless response (PMVDR) and Mel-frequency cepstral coefficients (MFCC). We generate WDFTC and PMVDR features by all-pass based warping; we use spectral warping for MFCC. PMVDR and WDFTC use warped-LP and warped discrete Fourier transforms, respectively. We employ WDFTC, PMVDR and MFCC features in continuous noisy monophone and word recognition tasks using the TIMIT corpus. We also test these features on gender-specific monophone and word recognition tasks. Further, we employ spectral amplitude warping (SAW) in WDFTC feature extraction (WDFTC SAW) and demonstrate enhanced robustness of this feature. We observe that SAW does not improve robustness for the MFCC and PMVDR features. Finally, we report the recognition performance and discuss many interesting properties of these features. Our study shows that the PMVDR and WDFTC SAWachieve recognition performance superior to the MFCC and WDFTC in noisy conditions.

Pages: 97 to 108

Copyright: Copyright (c) to authors, 2009. Used with permission.

Publication date: June 7, 2009

Published in: journal

ISSN: 1942-261x