Perception of synthetic silent-center syllables: A replication and extension

Abstract

The purpose of this study was to replicate and extend the synthetic silent-center vowel perception study of Miranda and Strange [J. Acoust. Soc. Am. Suppl. 1 88, S55 (1990)]. The goal was to generate synthetic syllables that adequately captured the dynamic spectral information in syllable onsets and offsets. Natural and synthetic /dVd/ syllables were presented to subjects in two forms: full syllable and silent center. Natural silent centers were constructed ty retaining the burst and first three and final four pitch periods in order to achieve a desired vowel identification error rate of about 10%. Natural duration was not altered. From ILS analysis of the first three formants, synthetic syllables were generated with Haskins Labs SYN software synthesis. Natural and synthetic full syllable error rates were 2%. Natural silent center error rates were 11%. Synthetic silent centers were 23%, far less than the 60% reported by Miranda and Strange. Thus these synthetic syllables appear to have captured some of the dynamic spectral information in the syllable onsets and offsets. [Work supported by NIDCD.]