4th International Conference on Spoken Language Processing

Philadelphia, PA, USA
October 3-6, 1996

Word Graph Rescoring Using Confidence Measures

Pablo Fetter (1), Frédéric Dandurand (2), Peter Regel-Brietzmann (1)

(1) Daimler-Benz AG, Research and Technology, Ulm, Germany
(2) McGill University, Department of Electrical Engineering, Montreal, Canada

This paper presents a novel approach to using confidence scores for word graph rescoring. For each word in the system's vocabulary, we computed the probability that the observation is correct given its acoustic score. Afterwards, we used these probabilities for rescoring word graphs outputted by the recognizer. We will present some implementation details as well as accuracy improvements obtained using this method.

Full Paper

Bibliographic reference.  Fetter, Pablo / Dandurand, Frédéric / Regel-Brietzmann, Peter (1996): "Word graph rescoring using confidence measures", In ICSLP-1996, 10-13.