Audio-visual speech recognition compared across two architectures.