This paper addresses the problem of automatic numeric recognition and under
standing in spoken language dialogue. We show that accurate numeric underst
anding in fluent unconstrained speech demands maintaining robustness at sev
eral different levels of system design, including acoustic, language, under
standing and dialogue. We describe a robust system for numeric recognition
and present algorithms for feature extraction, acoustic and language modeli
ng, discriminative training, utterance verification and numeric understandi
ng and validation. Experimental results from a field-trial of a spoken dial
ogue system are presented that include customers' responses to credit card
and telephone number requests. (C) 2001 Elsevier Science B.V. All rights re
served.