Vowel Perception

Abstract

We investigated vowel perception by 15 subjects using the single-electrode cochlear implant used at the House Ear Institute in Los Angeles. Subjects were postlingually deaf adults having histories of unsuccessful hearing aid use and a minimum of 6 to 12 months experience with the implant. Eleven American English vowels spoken by a male talker were tape recorded, digitized, analyzed, and controlled for the experiments. The stimuli were audiorecordings of both natural and loudness-matched vowels. Subjects rated the dissimilarity of both the naturally spoken and the loudness-matched vowels, and performed identification of the latter. Two normal-hearing subjects served as controls for the dissimilarity tasks. Multidimensional scaling, hierarchical clustering, and percent correct identification analyses were used to help determine the perceptual features used by the subjects in their judgments. Generally, the normal-hearing subjects took advantage of second formant (F2) frequency information. The cochlear-implant users relied primarily upon fundamental (F0) and first formant (F1) frequency information and demonstrated difficulty in vowel identification. No major differences were noted for the natural versus loudness-matched vowels. F2 information, requisite for accurate vowel recognition, did not correspond to any of the perceptual dimensions discerned in the results obtained from implant subjects.