| 摘要 |
A method for processing speech recognition including a system for generating a first probability of the word or sequence that words in a speech utterance is rejectable and a second probability that the word or sequence of words in the utterance is acceptable. A new Selected Phoneme Rejection Grammar is constructed of phonemes and/or merged phonemes. The highest first probability is calculated by comparing the utterance to sequence of the Rejection Grammar phonemes and/or merged phonemes. A Main Grammar is implemented in which the highest second probability is calculated that the utterance is composed of words in the Main Grammar. The two highest probabilities are compared and the utterance is accepted or rejected depending on the higher of the two highest probabilities. If accepted a sensible response may be made as determined by the specific application. If it is more probable that the sequence is in the rejection grammar, the utterance is rejected and as appropriate response as determined by the specific application is returned. The Rejection Grammar constructed by the new method is of small size, comprising often thirty or fewer selected phonemes. However, even with the small sized Rejection Grammar it is practical for use with small and large Main Grammars.
|
