14th International Congress of Phonetic Sciences (ICPhS-14)
San Francisco, CA, USA
This work presents a series of strategies developed for a Natural Language Processing (NLP) module, which receives as input a sequence of words generated by a speech recognizer. These strategies have proved to be very effective in detecting and correcting the errors generated at the speech recognition stage. These strategies can be classified into three blocks: lexicomorphological and grammatical flexibility, metrics. The model has been implemented into a real system (automatic telephone operator system); the results generated for this paper have been taken after parsing a Spanish corpus of 1,719 requests to the operator.
Bibliographic reference. López-Soto, María Teresa / Quesada, José F. (1999): "Error detection and error recovery from speech recognition: language engineering strategies", In ICPhS-14, 1637-1640.