Audio-visual speaker identification using coupled hidden Markov models