In this paper we present a method for adapting an articulatory model to a new speaker from acoustic data only. The main goal of this method is to make acoustic-to-articulatory inversion a fully automatic process. Speaker-specificity is modeled by a two dimensional scale factor, which makes it more flexible than VTLN methods. Validation of the method is performed on three speakers by comparing the scale factors found on medical images of the vocal tract and those estimated. These experiments show that the method is accurate provided that the speech material covers a wide acoustic space. Two additional experiments on longitudinal acoustic data are presented, in order to study vocal tract evolution with age.