An Experimental Study of the Identification of Filtered Japanese Vowels

Japanese vowels [i], [e], [a], [o], and [u] recorded by an adult male and an adult female were identified under various conditions of high- and low-pass filtering. 1. Discontinuous functions of per cent correct identification of the vowels versus filter cut-off frequency were found to be in the identification of [i] (HP), [e] (HP), [a] (HP), [o] (LP), and [u] (HP and LP) of the male speaker, and in the identification of [i] (HP), [e] (LP), [o] (HP and LP), and [u] (HP and LP) of the female speaker. The results indicate that the discontinuities exist in more cases than those suggested by Chiba and Kajiyama. 2. Difference was found to be between the correct identification of vowels of the male speaker and those of the female speaker reflecting their vowel spectrum envelopes. In general, vowels of the male speaker were identified better than those of the female speaker under low-pass filtering conditions, while the vowels of the female speaker were identified better than than those of the male speaker under high-pass filtering conditions. 3. Difference was found to be between the identification of vowels distorted by different sharpness of filtering. (Attenuation: 55 dB/octave, 2×55dB/octave, and 3×55dB/octave.) 4. There was no statistically significant difference in per cent correct between the identification of the unfiltered vowels of both speakers intend for /i/, /e/, /a/, and /u/ and the unfiltered [o] of the male speaker, which had high identifiablity (per cent correct: 100 to 9 except for [o], 98 to 80). The per cent correct identification of unfiltered [o] of the female speaker (per cent correct: 85 to 46.) was significantly lower than that of the other four vowels. 5. There were sometimes significant differences between the estimation of clearness of correctly identified vowels although the per cent correct identification of the vowels did not show any significant difference between them. The results imply that there are some frequency regions in which the vowel spectra still contribute to the phonetic quality of Japanese vowels, which cannot be revealed by merely phonemic categorization of the stimuli