ESCA Workshop on Audio-Visual Speech Processing (AVSP'97)
September 26-27, 1997
This article reports an experimental study which aimed at evaluating the extent to which discrepancies in gender between a seen face and a heard voice affect the intelligibility of acoustically degraded speech presented audio-visually to naive subjects. Our results compare identification scores across two groups of subjects, depending on whether they had been first familiarized with one of the speakers' face and voice.
Our results do not support the hypothesis that familiarity with a speaker would increase the benefit of lipreading when speaker identity is provided by the two modalities. More surprising are our results from the group of subjects who did not have a chance to see/hear the female speaker before the test. A majority of them identified her face as that of a male when presented visually. A majority of them also identified her voice as male when presented auditorily under highly degraded conditions. However, a majority identified her as female through bimodal presentation! This result challenges the models of bimodal integration based on synergetic reinforcement.
Bibliographic reference. Schwippert, C. / Benoît, Christian (1997): "Audiovisual intelligibility of an androgynous speaker", In AVSP-1997, 81-84.