In case of an isolated vowel or a CV syllable, a close relation can be found between the phonetic quality of a certain phoneme and its acoustic feature. This relation, however, is not always preserved to the phonemes in an ordinary connected speech. In our previous study, it was found that the perception of a vowel in a connected speech was seriously impaired by the complete removal of its phonetic environment and that two syllables, one preceding and one following, were necessary and sufficient to provide a perceptual environment for their correct identification. From this point of view, an identification experiment concerning the dynamic aspect of speech perception has been performed on synthetic vowels in terms of phoneme boundary location. As a typical example of phonetic environment vowels, symmetrical nonsense words /uVu/ were chosen for the perceptual experiment. The formant patterns assigned to the point of closest approach to target were selected from a set of 16 points spaced equally on a straight line in the F_1-F_2 plane (Fig. 1). Synthetic stimuli were presented to listeners under two conditions, steady-state and /uVu/ condition. Listeners were asked for identifying each synthetic vowel under the steady-state condition and the middle vowel in each /uVu/ word with one of Japanese five vowels. It was found that the phoneme boundary locations of vowels under the /uVu/ condition shifted largely, about 4 to 5 steps in stimulus number, compared to those of vowels under the steady-state condition (Fig. 6, Fig. 7 and Table 2). The result suggests that there exists a certain function in human auditory system so as to compensate the formant frequency undershoot associated with vowel reduction.
抄録全体を表示