In adverse acoustic condition, speech intelligibility is dramatically improved when the speaker's face can be seen. Which components of the face best convey the visual information associated with speech gestures? We answer this question through a series of experiments where normal hearers had to identify three vowels [a,i,y] and six consonants [b, v, z, 3, R, 1] in French nonsense words of the form VCVCV, presented audiovisually under five conditions of background noise and five conditions of visual displays (the real face of the reference speaker, his lips alone, and three synthetic displays of facial components: lips, jaw and lips, whole face.)
Bibliographic reference. Goff, Bertrand Le / Guiard-Marigny, Thierry / Benoît, Christian (1995): "Read my lips... and my jaw! how intelligible are the components of a speaker's face?", In EUROSPEECH-1995, 291-294.