An ideal observer analysis of variability in visual-only speech

Normal-hearing observers typically have some ability to “lipread,” or understand visual-only speech without an accompanying auditory signal. However, talkers vary in how easy they are to lipread. Such variability could arise from differences in the visual information available in talkers’ speech, human perceptual strategies that are better suited to some talkers than others, or some combination of these factors. A comparison of human and ideal observer performance in a visual-only speech recognition task found that although talkers do vary in how much physical information they produce during speech, human perceptual strategies also play a role in talker variability.