Prediction and constraint in audiovisual speech perception