Kikuo Maekawa and Hideaki Kikuhi (National Institute for Japanese Language):
Corpus-based Analysis of Vowel Devoicing in Japanese


   In this paper, vowel devoicing in Standard Japanese is 
analyzed based upon the quantitative analyses of a large-scale 
speech corpus known as CSJ, or Corpus of Spontaneous Japanese 
that the authors and their colleagues have been compiling since 
1999. 


   The subset of CSJ that we will use for this paper contains 25 
hours of spontaneously uttered Japanese monologue (mainly the live 
recording of academic presentation speech and recording of 
layman public speech). In addition to the segmental labels that 
were applied for the whole body of the subset, prosodic labels 
were applied for about 7 hour. 


   The particular aim of the analyses is a quantitative 
evaluation of the factors that influence the probability of 
vowel devoicing, both language-external and language-internal. 


   Language-external factors include speaking rate, speakers' 
age, speakers' sex, and the degree of relaxation as measured by 
the mean occurrence of speakers' laughter. Language-internal 
factors include types of the preceding and following consonants, 
vowel height, interaction with various prosodic events like 
lexical accent, and boundary pitch movements.