AN ALGORITHM FOR THE AUTOMATIC DIFFERENTIATION BETWEEN THE SPEECH OF NORMALS AND PATIENTS WITH FRIEDREICHS ATAXIA BASED ON THE SHORT-TIME FRACTAL DIMENSION

In this paper, we describe an algorithm, based on acoustic pattern mat ching techniques, for providing an automatic, highly reliable distinct ion between normal and some kind of pathological speech (Friedreich's ataxia disease). For each utterance, the short-time fractal dimension parameter and; for comparison, the zero-crossing and energy ratio para meters are evaluated and used in the classification task by means of a dynamic programming procedure. Although all the parameters are able t o differentiate the two groups, the fractal dimension parameter seems to provide a more reliable pattern classification than zero-crossing a nd energy ratio. Finally, we point out that, to the discrimination pur pose, an accurate choice of the utterances to be pronounced by the sub jects is to be considered. (C) 1998 Elsevier Science Ltd. All rights r eserved.