TU Berlin

Quality and Usability LabReviewed Conference Papers

Page Content

to Navigation

Reviewed Conference Papers

go back to overview

Improving Posterior Based Confidence Measures Using Enhanced Local Posteriors
Citation key ketabdar2010a
Author Ketabdar, Hamed
Title of Book The Eighteenth European Signal Processing Conference
Year 2010
ISSN 2076-1465
Location Makuhari, Japan
Month aug
Publisher Eurasip
How Published full
Abstract In this paper, we propose a new technique for enhancing posterior probability based con?dence measures in ASR systems. Wepropose to enhanced local posterior estimates used in con?dencemeasurement process, in order to improve the overal con?dencescore in terms of ability to accept/reject a hypothesis. Posteriorbased con?dence measures are global scores obtained by accumu-lating local evidences. These local evidences are usually phoneposterior probabilities estimated in frame basis from speech sig-nal. Having better (more informative) local evidences can poten-tially lead to better con?dence measures. In [1, 2, 3], a methodfor enhancing local phone posterior estimates (evidences) has beenproposed. This method is based on integrating prior knowledge(such as phone duration, lexical knowledge) and temporal context in the local posterior estimation. We show that using enhanced localposteriors in the con?dence measurement process significantly andconstantly improves their ability to predict whether a hypothesis (atword or phone level) is correct or incorrect, as compared to usingregular local posterior estimates.
Download Bibtex entry

go back to overview


Quick Access

Schnellnavigation zur Seite über Nummerneingabe