Utterance Verification Based on Support Vector Machine

BF Yan,XY Zhu

2004-01-01

Abstract:In this paper we present an approach to recognition confidence scoring and techniques for integrating confidence scores to make final decision in speech recognition. The measure is useful for rejecting utterances that contain out-of-vocabulary words. We derive a feature vector consisting of the filler-based posterior probability parameters. An accept/reject classifier is trained to produce a single confidence score for a hypothesis. We train and test a support vector machine (SVM) classifier, then compare the results with other statistical classification methods in the experiments.