Integration of utterance verification with statistical language modeling and spoken language understanding

Methods for utterance verification (UV) and their integration into statisti cal language modeling and understanding formalisms for a large vocabulary s poken understanding system are presented. The paper consists of three parts . First, a set of acoustic likelihood ratio (LR) based UV techniques are de scribed and applied to the problem of rejecting portions of a hypothesized word string that may have been incorrectly decoded by a large vocabulary co ntinuous speech recognizer. Second, a procedure for integrating the acousti c level confidence measures with the statistical language model is describe d. Finally, the effect of integrating acoustic level confidence into the sp oken language understanding unit (SLU) in a call-type classification task i s discussed. These techniques were evaluated on utterances collected from a highly unconstrained call routing task performed over the telephone networ k. They have been evaluated in terms of their ability to classify utterance s into a set of 15 call-types that are accepted by the application. (C) 200 1 Elsevier Science B.V. All rights reserved.