Semi-automatic extraction of vocal tract movements from cineradiographic data
Julie Fontecave, Frédéric Berthommier · 2006
S ince high speed X -ray film s still provide the best dynam ic view of the entire vocal tract, large existing databases have been preserved and are available for the speech research com m unity. W e propose a new technique for facilitating the extraction of the vocal tract shape and the m ovem ents of the articulators from com plete sequences of these databases. T he m ethod w as first developed for the extraction of the tongue m ovem ents in W ioland. It has been adapted to a sequence of the A T R database, L aval43 (provided by R ochette). T he m ethod, based on the retrom arking algorithm , com bines the hum an expertise applied for m arking a sm all num ber of key im ages, and the autom atic processing of the video data. It has been extended to other articulators (lips, velum ) in order to obtain the shape and the sections of the com plete vocal tract. Q uantitative evaluations of the estim ate error and a com parison w ith T him m and L uettin (1999) are achieved. In d ex term s: sem i-autom atic extraction, cineradiographic data processing, X -ray film database, vocal tract contour extraction