MOTHER: MOrphable Talking Head for Enhanced Reality
The project aims at producing virtual clones of the head of human speakers
where the movements of all speech articulators produce coherent deformations
of both visible and unvisible organs. MOTHER is driven by a linear articulatory
model trained on experimental data gathered on the subject using various
exprimental setups: video for face data, cineradiography and MRI for tongue
data. This linear articulatory model generates coherent movements of feature
points of the meshes definig the different 3D articulators (skin, lips,
tongue, teeth, jaw...). The parameters listed below
define an additionnal MPEG4 layer called TAP (Task-specific Articulatory
Parameters) over the FAP layer.
Applications are numerous:
-
Analysis-Compression-Synthesis of talking faces (cf. the Labiophone
project) for telecommunications and videoconferencing
-
Text to Audiovisual Speech System (see demo)
-
Language learning & phonetic correction
References
-
Lionel Réveret, Gérard Bailly & Pierre Badin; MOTHER
: a new generation of talking heads providing a flexible articulatory control
for video-realistic speech animation, Proceedings of ICSLP, Beijing-China,
2000. (PostScript,
4 pages, 1574688 bytes)
Demos
-
Articulatory degrees-of-freedom for speech
-
Composing the articulatory degrees-of-freedom to mimick
facial movements
-
Text-to-Audiovisual Speech System
icp.avi:
talking head with underlying skull and jaw made visible
Retour à la page de G. Bailly
Retour à l'index de l'ICP
Last
modified: Thu Feb 11 13:44:53 MET 1999