Algorithm for n-best ASR result processing to improve accuracy

摘要

A method of improving speech recognition includes receiving a plurality of recognized sentences from an automatic speech recognition (ASR) engine, each of the plurality of the recognized sentences including a plurality of information items. A confirmation is also receive that a value of an information item of the plurality of information items has been misrecognized or recognized. If the value has been misrecognized, each of the plurality of sentences are deleted that include the value of the information item that was misrecognized. If the value has been recognized, each of the plurality of sentences that do not include the value of the information item is deleted. The remaining sentences are transmitted to a dialog manager module.