Comparing support vector machines, recurrent networks, and finite state transducers for classifying spoken utterances

Abstract

This paper describes new experiments for the classification of recorded operator assistance telephone utterances. The experimental work focused on three techniques: support vector machines (SVM), simple recurrent networks (SRN) and finite-state transducers (FST) using a large, unique telecommunication corpus of spontaneous spoken language. A comparison is made of the performance of these classification techniques which indicates that a simple recurrent network performed best for learning classification of spontaneous spoken language in a robust manner which should lead to their use in helpdesk call routing. © Springer-Verlag Berlin Heidelberg 2003.