This study was undertaken to examine relationships between acoustic speech measures and auditory phonetic perception. The hypothesis of the study was that physical dissimilarity of the acoustic measures could substantially account for perception. Speech samples of 22 Consonant-/a/ syllables and 14 /h/-Vowel-/d/ syllables were spoken by two talkers. The stimuli were processed by vocoders with two different filterbanks. Forced-choice perceptual identifications were obtained from 6 normal hearing participants for each talker, vocoder, and syllable set. Confusion data were analyzed using multidimensional scaling, and Euclidean distances among stimulus phonemes were computed. For each pair of stimuli (within the same talker, vocoder, and syllable set), physical Euclidean distances were computed within frequency channels and averaged across time. Multilinear regression was used to transform the Euclidean physical distances to perceptual distances. Evaluation using Pearson r showed that the transformed physical distances correlated with perceptual distances between 0.55 and 0.96 (30% to 91% variance accounted for), depending on the talker, vocoder, and syllable set. The results indicated that the distinctiveness of the speech signals can account for perceptual dissimilarity structure by using a linear transformation.
Cite as: Xue, J., Takayanagi, S., Bernstein, L.E. (2002) Accounting for perceptual identification of consonants and vowels through acoustic dissimilarity. Proc. 7th International Conference on Spoken Language Processing (ICSLP 2002), 1649-1652, doi: 10.21437/ICSLP.2002-489
@inproceedings{xue02_icslp,
author={Jianxia Xue and Sumiko Takayanagi and Lynne E. Bernstein},
title={{Accounting for perceptual identification of consonants and vowels through acoustic dissimilarity}},
year=2002,
booktitle={Proc. 7th International Conference on Spoken Language Processing (ICSLP 2002)},
pages={1649--1652},
doi={10.21437/ICSLP.2002-489}
}