Prediction of listener perception of place-of-articulation in reduced speech

Abstract

Previous research on stop consonant production found that less than 60% of the stops sampled from a connected speech corpus contained a clearly defined hold duration followed by a plosive release [Crystal and House, JASA(1988)]. How listeners perceive reduced, voiced stop consonant variants is not well understood. The purpose of the current study was to investigate whether an acoustic cue called a relative formant deflection pattern was capable of predicting listeners’ perceptions of these approximant-like, voiced stop consonants variants. A new methodology motivated by a computational model of speech production was used to extract relative formant deflection patterns from excised VCV segments from a reduced speech database. Participants listened to a total of 56 excised VCV stimuli containing approximant-like, voice stop consonant variants and performed a force choice test (i.e., /b-d-g/). The agreement between the perceptions predicted by the relative formant deflection patterns and listeners’ behavioral performance was compared. The expected relative formant deflection pattern correctly predicted listeners' primary response for percent /b/ and /g/ identifications, but not for listeners’ percent /d/ identifications. The implications of these results on a possible invariant acoustic correlate for listeners’ perceptions of place-of-articulation information will be discussed.