Minimum spectral contrast for vowel identification by normal and hearing-impaired listeners
Marjorie R. Leek, Michael F. Dorman, Quentin A. Summerfield · The Journal of the Acoustical Society of America · 1985
It has been suggested that the internal spectral representation of vowels in hearing-impaired listeners contains considerably less information than is available in the acoustic speech signal. It is puzzling, then, that unless the loss is quite severe, such listeners typically have little trouble identifying vowels. Apparently, faithful preservation of the vowel spectrum is not necessary for accurate identification. In this study, we sought to determine the minimum peak-to-valley differences in formant amplitude which could adequately define particular vowels for normal-hearing and hearing-impaired listeners. Vowel-like complex tones were generated by adding 30 harmonics of a 100-Hz tone in cosine phase. Amplitudes of all harmonics were equal, except for six components selected to define the formants of four vowels, which listeners were asked to identify. The selected components varied relative to the remaining component amplitudes by ±1 to +8 dB. Results indicated that the 2–3-dB peak-to-valley contrast required for identification by normal hearing listeners had to be doubled and tripled, respectively, for normal listeners in noise and for hearing-impaired listeners. While the spectral contrast information in vowels is more than sufficient for identification by the hearing-impaired, these listeners are operating closer to the limits of correct identification than their normal-hearing counterparts. This procedure offers a paradigm for a finer quantification of differences in vowel perception by these two populations than the usual accuracy measures. [Research supported by NIH.]