fIntegrating confidence scores for utterance verification

In this paper we present an approach to integrating confidence scores for utterance verification based on neural network. In addition to the confidence scores computed at the phonetic level, we use various novel confidence scores, including segmental confidence scores, likelihood ratio and recognition results of LPC feature. We describe a method to combine different confidence scores via a neural work for word hypotheses. Our experimental evaluation shows that the novel confidence scores are effective and the proposed combination method gives better results than other classification method in keyword spotting system.