RELATIONAL ACOUSTIC INVARIANCE IN JAPANESE VOWEL LENGTH DISTINCTION

要旨

The issue of acoustic invariance (Pickett et al., 1999) was explored with the question of whether a relative duration value can be found to reliably classify the Japanese phonemic short and long vowels produced across three speaking rates. Four native Japanese speakers read ten pairs of disyllabic real words in a carrier sentence at slow, normal, and fast rates three times each. The duration of short vowels in CVCV contexts and contrasting long vowels in CVVCV and CVCVV contexts, as well as word duration, was measured. The short vowels spoken at the slow rate and the long vowels spoken at the fast rate overlapped significantly in duration. Rate changes affected the duration of long vowels more than short vowels, and the duration difference between short and long vowels was greater for slower speech. In contrast with absolute duration, the ratios of the target vowel to the total word duration (means = short: 0.29; long: 0.46) did not change across rates. The optimal boundary ratio that distinguished the two vowel length categories across three rates was found to be 0.38, and this ratio was identical to that obtained in Japanese nonwords with controlled segments in /mV(V)mV(V)/ contexts (Hirata, 2003). The results support the view that relational acoustic invariance exists in the speech signal that remains stable across speakers and rates.

主題