Facial expression analysis and auditory perception

The psychoacoustic assessment of loudness disorders like Hyperacusis relies on subjective reports, which are biased and impractical for non-verbal patients. Similar issues in pain evaluation have been addressed using facial expression analysis to decode the intensity and affective value of perceived pain. Thus, we are developing a system for the objective evaluation of perceived sounds from a listener’s facial expression. We determine whether sensory (intensity) and affective (valence) dimensions of sound perception can be distinguished through facial expressions by employing action unit analysis and other facial feature extraction methods on our in-house dataset. This video dataset includes facial expressions in response to sounds with varying intensities and emotional valences. We train convolutional networks to decode these dimensions from both extracted features, such as AUs and raw video data. By employing feature and decision fusion methods, we are developing an automated system for the objective assessment of perceived sound. This system will pioneer the first objective method for assessing loudness disorders, enhancing patient care and potentially distinguishing between hyperacusis sub-types to facilitate personalized treatment.