HMM SPEECH RECOGNIZING SYSTEM, CHARACTERISTIC SAMPLING DEVICE AND METHOD FOR RECOGNIZING, RECOGNIZING CONTROL METHOD, CONSTRUCTING METHOD OF NEW VOCABULARY PRONUNCIATION DICTIONARY

摘要

A speech recognition system using Hidden Markov Model (HMM) and the controlling method therefor is provided to enhance recognition time by extracting characteristics of parameters of HMM such that the system differentiates the characteristics in the recognition process from that in the extracting process. The speech recognition system comprises a speech dictionary building means(6) for changing the a speech dictionary to a recognizing speech dictionary; an HMM parameter storing means(7) in connected with the speech dictionary building means(6) for storing HMM parameters; a new vocabulary speech dictionary building means(5) in response to the input of a new vocabulary to provide the speech dictionary building means(6) with a speech dictionary for the new vocabulary; a recognizing characteristics extracting means(4) for extracting the characteristics for use in obtaining the HMM parameter relative to the speech input; and an HMM recognizing means(8) for outputting results of the speech recognition based on the characteristics extracted from the extracting means(4) and the speech dictionary in the speech dictionary building means(6).