In Proceedings of the European Signal Processing

Abstract

In this paper we propose a new approach to integrating a parser into a statistical speech recognizer. The method is able to incrementally apply grammatical restrictions and robustly combine them with the statistical acoustic and language models. On spontaneous speech data a 11.6% reduction in word error rate could be achieved compared to the baseline system applying statistical models only.

Extracted Key Phrases

Cite this paper

@inproceedings{Wachsmuth1998InPO, title={In Proceedings of the European Signal Processing}, author={Sven Wachsmuth and Gernot A. Fink and Gerhard Sagerer}, year={1998} }