Automatic detection and evaluation of edentulous speakers with insufficient dentures

Abstract

Dental rehabilitation by complete dentures is a state-of-the-art approach to improve functional aspects of the oral cavity of edentulous patients. It is important to assure that these dentures have a sufficient fit. We introduce a dataset of 13 edentulous patients that have been recorded with and without complete dentures in situ. These patients have been rated an insufficient fit of their dentures, so that additional (sufficient) dentures and additional speech recordings have been prepared. In this paper we show that sufficient dentures increase the performance of an ASR system by ca. 27 %. Based on these results, we present and discuss three different systems that automatically determine whether the dentures of an edentulous person have a sufficient fit or not. The system with the best performance models the recordings by GMMs and uses the mean vectors of these GMMs as features in an SVM. With this system we were able to achieve a recognition rate of 80 %. © 2010 Springer-Verlag Berlin Heidelberg.