Recognizing emotions from speech using a physical model