Stop consonant discrimination based on human audition

Abstract

A system for discrimination of stop consonants has been designed on the basis of studies of auditory physiology and psychophysics. The system consists of a one‐third octave filter bank as an approximation to auditory tuning curves, a bank of high speed, wide dynamic range envelope detectors, a logarithmic amplifier, and a digital computer for analysis and display. Features, chosen on the basis of psychophysical experiments, are then abstracted, and fed to a discriminant analysis program which decides on the most probable phoneme. Discrimination accuracy of about 77% for stop consonants in initial position has been achieved, with a 15‐speaker data set.