Notes on so-called inter-speaker difference in spontaneous speech: the case of Japanese voiced obstruent

In spontaneous speech, seemingly ‘inter-speaker’ variation is not thoroughly speaker-dependent. It is often the cases that, in addition to the genuine speaker-dependent factors, contextual factors like speaking rate and linguistic environment also have their influence on the variation. In this paper, the variations in the vocal-tract closure articulation of Japanese voiced obstruent consonants are reanalyzed in view of the separation between speaker-dependent and contextual factors. The consonants in the Corpus of Spontaneous Japanese were analyzed by means of logistic regression using the realization of vocal-tract closure as the binary dependent variable and the TACA (time allotted for consonant articulation), a recently proposed articulatory parameter, as the independent variable. Thresholds of TACA for the realization of vocal-tract closure were computed for all speakers. The results revealed clear inverse correlation between the threshold and the mean realization rate of closure. To explain the perturbation in the correlation pattern, influence of skewed distribution of phonemes and pauses were evaluated. The results showed the expected patterns in /z/ and /b/, but not in /d/ and /g/.