Effects of speaking rate on the single/geminate stop distinction in Japanese.

This study addressed an issue in the theory of acoustic invariance. The question was whether an invariant acoustic property exists for distinguishing Japanese single and geminate voiceless stops across rates and speakers. Four native Japanese speakers produced disyllabic words with single and geminate stops (e.g., /kako/ and /kak:o/) spoken in a carrier sentence at three speaking rates. Durations of sentences, words, stop closures, vowels preceding the contrasting stops, and voice onset times were measured. Ratios of geminate to single stop closures, geminate words to singleton words, closures to preceding vowels, and closures to words were calculated. The effect of rate on closure duration was to yield overlap between the singleton and geminate categories, and to lengthen geminate closures more than single closures as rate decreased. However, the ratio of geminate to single closure duration was unaffected by rate. Furthermore, the ratio of closure to word duration (0.35 as an optimal boundary) best classified all singleton and geminate tokens with 95.7%-98% accuracy. Thus, in spite of overlap in absolute closure duration between single and geminate voiceless stops, there is a relationally invariant measure that divides the two phonemic categories across rates and speakers, supporting the theory of relational acoustic invariance. co ustical Society of America.