1st ETRW on Speech Production Modeling: From Control Strategies to Acoustics 4th Speech Production Seminar: Models and Data Autrans, France May 20-24, 1996

An Investigation of Hypo- and Hyper-Speech in the Visual Modality

Christian Benoît (1), A. Fuster-Duran (2), B. Le Goff (1)

(1) Institut de la Communication Parlée, UPRESA CNRS n° 5009, INPG/ENSERG, Université Stendhal, Grenoble, France
(2) Institut für Phonetik, Universität zu Köln, Köln, Germany

Is visible speech more or less intelligible when a face is hyper-articulated or animated with standard motion at a given speech rate? Visual speech intelligibility was compared across two conditions of articulation of a parametric face model. An audio-visual-speech synthesizer was used to generate visual stimuli at two different rates, both with hypo- and with hyper-articulation. Hypo-articulation at the conversational rate was obtained by increasing coarticulation so that trajectories of the command parameters matched that obtained in the fast rate. And vice-versa for hyper-articulation. Results must be interpreted with care since the synthesizer used is still in its early age. Although all differences are not significant, results tend to show that speechreading is at its best when articulation is standard. Hypo-articulation is less intelligible. Hyper-articulation also seems to be less intelligible, but this last result remains to be confirmed.

Full Paper

Bibliographic reference.  Benoît, Christian / Fuster-Duran, A. / Goff, B. Le (1996): "An investigation of hypo- and hyper-speech in the visual modality", In SPM-1996, 237-240.