In this article, we present first conclusions about labial coarticulation strategies drawn from a corpus of speech audiovisual data. The general idea is to develop a talking head which would be understandable by lip readers especially deaf persons. With a stereovision system, we recorded a corpus with ten French native speakers (5 female and 5 male speakers). Visual and audio information was analysed to extract the labial parameters (opening of the lips, stretching and protrusion). Even if the analysis shows a great variability between speakers, we nevertheless found general tendencies which would help us to develop a reliable prediction algorithm of labial coarticulation.