Abstract: In this paper we present a new technique to enhance the performance of spoken dialogue system, which employs contextual models and grammatical rules to optimise the automatic correction of some errors made by the ASR component of these systems. Different experiments have been carried out to evaluate this technique employing a previously developed spoken dialogue system designed for the fast food domain. The results of this experimentation show that our technique enhances the performance of the system, by notably incrementing the rates of word accuracy, speech understanding and task completion.
Index Terms: Spoken Dialogue Systems, Automatic Speech Recognition, Language Modelling.