Speech Prosody 2002
Japanese is a tonal language, and it is also mora-timed language. In this paper, we investigated the hypothesis that accent kernels appear at regular intervals. If duration of each mora is almost equal, then Japanese accent kernel is a constant time marker of continuous Japanese. The Japanese prosodic corpus we investigated comprises 3 hours and 37 minutes of speech involving six different speakers (3 males and 3 females). The recordings on which the corpus is based are passages translated from the MULTEXT prosodic corpus distributed through the European Language Resource Association (ELRA). The text was translated from five European languages and modified into Japanese, consisting of 40 different passages, with 6523 morae including 1085 accent kernels. Speakers are professional narrators and actors who read the text naturally before recording. During the recording, the standard Japanese accent was kept. Accent kernels are marked by two experienced researchers of Japanese phonology. ANOVA assured the average occurrence of accent kernel is one out of every six morae.
Bibliographic reference. Kitazawa, Shigeyoshi / Kitamura, Tatsuya / Mochizuki, Kazuya / Itoh, Toshihiko (2002): "Periodicity of Japanese accent in continuous speech", In SP-2002, 435-438.