EXTRACTING KINEMATIC PROPERTIES FROM LARGE-SCALE ULTRASOUND CORPORA

International audienceThis study describes a highly automated approach to quantifying articulatory movements in ultrasound echography, and offers an example of its application in the examination of the timing of consonant-vowel coordination in typologically contrasted languages (French and Japanese). Despite recent developments in data acquisition and storage facilities, ultrasound requires a timeconsuming post-process consisting of e.g., semimanual tongue contour tracking, that prevents it from being fully involved in large-scale phonetic corpora. To take over this limitation, our approach looks at frame-to-frame pixel-wise variation of brightness. The articulatory targets can then be expressed as local minima of the rate of change. Preliminary results from a bilingual speaker suggest a difference in the time coordination of consonantal and vocalic targets in French and Japanese before and after the intervocalic /k/