Increasing Robustness of Spoken Dialogue Systems » książka
Increasing Robustness of Spoken Dialogue Systems
ISBN-13: 9783836476782 / Angielski / Miękka / 2009 / 104 str.
In conversational dialogue applications it iscritical to understand the requests accurately.However, the performance of current speechrecognition systems are far from perfect. In order tofunction effectively with imperfect speechrecognition, an accurate confidence scoring mechanismshould be employed. To determine a confidence scorefor a hypothesis, certain confidence features arecombined. In this work, the performance offiller-model based confidence features areinvestigated. Five types of filler model are defined:triphone-network, phone-network, phone-class network,5-state catch-all model and 3-state catch-all model.First all models are evaluated in terms of theirability to correctly tag (miss or hit) recognitionhypotheses. Then the performance of reliablecombinations of these models are evaluated to showhow certain reliable combinations of filler modelscould significantly improve the accuracy of theconfidence annotation. Moreover to show the practicalside of the work, an implementation of a realdialogue management system is described.
Czytaj nas na:
KrainaKsiazek.PL - Księgarnia Internetowa
