Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

A Comparative Study of Feature Representations for Robust Speech Recognition in Adverse Environments

Kuldip K. Paliwal, Bishnu S. Atal

Speech Research Department, AT&T Bell Laboratories, Murray Hill, NJ, USA

In this paper, a number of feature representations are studied as to their recognition performance in presence of additive noise and channel mismatch distortions. It is shown that 1) the linear prediction analysis technique provides more robust cepstral features than the homomorphic analysis technique, 2) the filter-bank power spectrum is capable of generating more robust cepstral features than the power spectrum derived through the fast Fourier transform algorithm, and 3) use of human auditory properties in an acoustic front-end makes it more robust.

Full Paper

Bibliographic reference.  Paliwal, Kuldip K. / Atal, Bishnu S. (1994): "A comparative study of feature representations for robust speech recognition in adverse environments", In ICSLP-1994, 1015-1018.