Uncertainty Handling at Speech/Natural Language Interface. Volume 1.

摘要

VoiceScribe 1000 Speech Recognition System provides a discrete utterance, speaker dependent recognition system capability. The speech driver interface routines give a 'confidence level,' which is a number between 0 and 100, for the word that best matches the uttered word. For the alternate choices it provides distances between those words and the uttered word, which are numbers between 0 and 254. To modify the VoiceScribe 1000 Speech Recognition System to give not only the best match, but a set of close matches for each utterance and their levels of match, the distance measures were utilized to calculate the certainty levels for alternate choices, and the confidence level for the best matched word was used directly as its certainty level. Further work needs to be done for more consistent confidence level assignments.