摘要：The automatic speech understanding aims to extract the useful meaning of the oral
utterances. In this paper, we propose an original method for automatic Arabic speech understanding.
The proposed method is strongly guided by the semantics. It has the advantage of being robust when
coping with the disfluencies such as the hesitations, the repetitions and self-corrections. We also
present, the SARF system, an interactive Arabic vocal server for Tunisian railway information. The
speech-understanding module of SARF is based on the proposed method. The evaluation results of
this module are very encouraging. Indeed, the obtained rate of F-Measure is 71.79%.
