分割された日本語単音節の知覚実験的研究

Abstract

The purpose of this study was to investigate the possible influence of a certain method of segmentation of speech sounds upon their identification. Procedure : 1) A male utterance of forty Japanese discrete CV syllables, which included fifteen consonants [p, t, k, b, d, g, s, h, dz, d, r, w, j, m, n] and five vowels [a, i, u, e, o], was recorded on magnetic tape. 2) Positional matching between a syllable's recorded location on the tape and its indicated position on a spectrogram was done by means of a technique employing a sound spectrograph. 3) A boundary between consonants and vowels was determined at the point where the vowel formants began to enter a quasi-stable state on the spectrogram. 4) The tape was cut vertically at two points each of which respectively bisected the vowel duration and the consonant duration (in the case of the voiced consonant, for this study, the buzz bar was not considered to be within the consonant portion). 5) The initial half of consonant segment tapes and the latter half of vowel segment tapes were joined together to make new syllables named "initial consonant syllables" from the consonant portion weighting and referred to as "IS" hereafter (Figure 1). 6) The remaining halves of the consonant segments and vowel segments were then left to form "transitional syllables", referred to as "TS" hereafter (Figure 1). Thus two kinds of syllables having equal duration in details were synthesized from one original syllable (Figure 2). 7) These syllables were edited in random order and re-recorded to make a test tape. 8) This test recording was then played to a listening panel of twenty students in psychology. Result : 1) Syllable intelligibilities for the IS were significantly higher in general than for the TS (Table 1). 2) Vowel intelligibilities for both IS and TS were affected only slightly from bisecting the durations (Table 2). The influence of bisecting, however, was somewhat more marked in the vowels [i] and [u] than in the other vowels (Table 3). 3) Consonant intelligibilities were approximately as high as for whole syllables, because vowels in both IS and TS were highly intelligible (Tables 1 and 4). Aside from the intelligibilities of consonants, a study of consonant perception using a confusion matrix revealed the following results (Table 4). Stop consonants among IS were highly intelligible, but for TS they were frequently not heard, especially the consonants [p] and [b]. Fricatives and affricatives in IS were also highly intelligible. For TS ; [s] sounds were heard most frequently as [t], while identifications of [k] for [h] stimulus sounds and of [d] for [d] stimulus sounds were dominant among error responses. In the hearing of liquids, misidentifications in IS increased as much as in TS. In both IS and TS, mishearing of [r] and [b] was marked. For other liquids, the observation of misidentification tendencies was not reliably established. For IS the perception of nasals as nasals was most frequent, while in TS correct identifications and [b]-hearings were dominant. For each consonant sound the possible perceptual cues were analyzed for this paper by means of spectrographic studies of consonant sounds and by studies of confusion matrices. From this data, it was concluded that initial durations contained a higher concentration of phonetic information about consonants than did final durations.