In this Diplomarbeit, different methods for visual tracking of lip motion in video sequences showing the head of a talking person shall be implemented and compared. Methods from the literature, such as deformable contours (Snakes), shall be extended to allow the extraction of features that are of particular relevance for further phonetic analysis. The developed system is intended to be a tool for researchers in linguistics. The features can be used as input parameters to automatic speech recognition systems and complement the parameter vectors derived from the microphone signal, they can be used to improve the implementation of existing coarticulation models and, at a later stage, also for the visual synthesis of lip motion for arbitrary spoken language.
This Diplomarbeit is a joint project between the Computer Graphics Group at the Max-Planck-Institut für Informatik and the Institute of Phonetics at the Universität des Saarlandes.