Speech recognition system having speech registration function based on twice utterance of word.

摘要

In a speech recognition system where a speech analyzer (12) analyzes an input utterance to produce an input speech pattern which is memorized in an input memory (14) as a memorized input pattern, a pattern comparator (16) comparing the memorized input pattern with reference patterns previously memorized in a reference pattern memory (15) and producing pattern dissimilarities between them, and a speech deciding circuit (18) deciding the minimum pattern dissimilarity to thereby recognize the input utterance, a word is uttered twice during a registration mode in order to correctly register the speech pattern of the word the reference pattern memory (15). Speech patterns of the first and the second utterances of the word are memorized in the reference pattern memory (15) and the input memory (14) as first and second memorized patterns, respectively, which are compared with each other at the pattern comparator (16) to produce a pattern dissimilarity. A registration deciding circuit (19) is selectively coupled to the pattern comparator (16) during the registration mode and decides whether or not the pattern dissimilarity is larger than a predetermined reject value and permits one of the first and the second memorized patterns to be registered as a reference pattern into the reference pattern memory (15) when the pattern dissimilarity is not larger than the reject value.