Effects of speaking rate on the perception of phonemic length contrast in Japanese

Abstract

Segment length is distinctive in Japanese, for example, /kaite/ (buyer) versus /kaite:/ (seabed). Such length contrasts are not necessarily categorical for non-native speakers. To study this property precisely, a series of perception experiments was conducted. A professionally trained native-Japanese speaker produced the nonsense word /erete/ at slow, normal, and fast rates with or without a carrier sentence. Either the second vowel or second consonant of each word was gradually lengthened until reaching its longer counterpart, i.e., /ete:te/ or /eret:e/, in all rate and carrier conditions using STRAIGHT, a high-fidelity speech analysis, synthesis, and manipulation system [Kawahara et al., Speech Commun. 27, 187–207 (1999)], resulting in 12 stimulus continua. Seven native-Japanese listeners participated in a single-stimulus, two-alternative forced-choice identification task with the method of constant stimuli. The speaking rate of the presented stimuli within a session was either fixed or randomized trial by trial. Results suggest that native listeners’ identification boundaries systematically altered due to changes in speaking rate, whereas their boundaries became unstable in the randomized-rate condition, especially for no-carrier stimuli. These results will be discussed from the viewpoint of second-language phoneme perception and acquisition through comparisons with results from non-native listeners. [Work supported by TAO, Japan.]