In-quiet and Lombard audiovisual speech

This experiment examined whether the visual speech enhancement in the intelligibility of degraded acoustic speech is greater when the speech signals were recorded in noise compared to when they were recorded in quiet. Ten sentences were spoken by four people either in-quiet or whilst they listened to cocktail party noise. The visual speech of these talkers was measured using 24 facial markers and associated video clips filmed. Sixty participants were tested in a speech-in-noise identification experiment under four different conditions. Visual speech enhancement was measured by mean percent words correctly identified in the audiovisual minus auditory-only condition scores. After ceiling effects were curtailed the results showed that the audiovisual enhancement for speech signals recorded in noise (43%) was significantly greater than that for speech recorded in quiet (30%).