Error Handling in Spoken Dialogue Systems August 28-31, 2003 Château d'Oex, Vaud, Switzerland

Recognition Error Handling by the Speech Understanding System to Improve Spoken Dialogue Systems

Caroline Bousquet-Vernhettes, Nadine Vigouroux

IRIT, Université Paul Sabatier, Toulouse, France

The aim of this paper is to increase the robustness of speech understanding facing the different kinds of recognition error. Indeed, these errors are frequent in human-machine dialogue and have usually repercussions on all components of the spoken dialogue system.

We propose an approach based on a stochastic and conceptual model. This model is robust to some kinds of errors like insertions and deletions. We present also an extension of this model in order to increase the robustness of the understanding process faced with misrecognitions (i.e. substitutions) and unknown words. We performed trial series on train schedule inquiry application to evaluate the understanding rate when confronted to recognition errors.

Full Paper

Bibliographic reference.  Bousquet-Vernhettes, Caroline / Vigouroux, Nadine (2003): "Recognition error handling by the speech understanding system to improve spoken dialogue systems", In EHSD-2003, 113-118.