Audiovisual-to-articulatory speech inversion using Active Appearance Models for the face and Hidden Markov Models for the dynami