2D Deformable Models for Visual Speech Analysis