Agent virtuel signeur - Aide à la communication des personnes sourdes

We present in this Phd report our investigation dedicated toward the de-sign and animation of autonomous and realistic signing agents. The highlightis focused on expressive quality of gestures.We show how expressive quality of gesture may convey meaningful infor-mation in the context of iconicity.We then propose a discourse model that is able to handle both illustrativeand non illustrative intends involved in sign language production.Tanks to motion analysis performed on a gesture sequence that is capturedon a professional deaf sign language instructor, we highlight key expressiveparameters.We then propose motion generation methods that suits our gesture speci-fication model. this motion generation method rely on several gesture genera-tion methods that are able to handle the variability induced by the requestedexpressive quality of the gesture.