ISCA Archive AVSP 1997
ISCA Archive AVSP 1997

Interpreted multi-state lip models for audio-visual speech recognition

Michael Vogt

This paper presents the application of a specialized language for the task of lip modeling. Models presented here are switching their state during evaluation. State changes are based on a highly accurate neural network decision. Different model states include optional parts for the inner mouth region. Using the contour description and a neural coded color profile, encouraging performance figures have been achieved.