European Conference on Speech Technology

Edinburgh, Scotland, UK
September 1987

Automatic Speech Recognition and Human Auditory Perception

Hynek Hermansky

Speech Technology Laboratory, Santa Barbara, California, USA

Low-order all-pole model of critical-band auditory spectrum, the Perceptually based linear predictive (PLP) model, compared with the standard linear predictive (LP) model, yields higher recognition accuracy in speaker-independent automatic speech recognition (ASR) while offering significant computational savings. Paper reviews the principle of the PLP method, shows and discusses some ASR results, and experimentally evaluates and discusses relationships of the PLP parametric representation of speech to several theories of speech perception. Results support use of model of speech perception in ASR front-ends.

Full Paper

Bibliographic reference.  Hermansky, Hynek (1987): "Automatic speech recognition and human auditory perception", In ECST-1987, 1079-1082.