EUROSPEECH '91
Second European Conference on Speech Communication and Technology

Genova, Italy
September 24-26, 1991

                 

Compensation for the Effect of the Communication Channel in Auditory-Like Analysis of Speech (RASTA-PLP)

Hynek Hermansky (1), Nelson Morgan (2), Aruna Bayya (1), Phil Kohn (2)

(1) US WEST Advanced Technologies, Boulder, Colorado, USA
(2) International Computer Science Institute, Berkeley, California, USA

Human perception is generally insensitive to steady-state stimuli, responding largely to relative values of the input stimulus. In spite of that, most speech parameter estimation techniques use absolute spectral values. Consequently, they fail when the spectral values are modified by the frequency response of the communication channel. In response to this challenge, we have developed an analysis technique that is more robust to steady-state factors in speech. The approach appears to work well with realistic speech and channel deformations, and is conceptually simple and computationally efficient. We have conducted an experiment on recognition of telephone-quality digits with a distorted channel. The new method is described, and experimental results for realistic channel distortion are reported, showing order-of-magnitude improvements in error rate.

Full Paper

Bibliographic reference.  Hermansky, Hynek / Morgan, Nelson / Bayya, Aruna / Kohn, Phil (1991): "Compensation for the effect of the communication channel in auditory-like analysis of speech (RASTA-PLP)", In EUROSPEECH-1991, 1367-1370.