PURPOSE: To improve the recognition rate, to reduce the restrictions of speaking on a user and accept a free speech, and to expand the utilization range by recognizing a variation of marry word series included in the free speech.
CONSTITUTION: This method is a continuous speech recognizing method which recognizes a continuously spoken speech by continuously matching the input speech with word standard patterns whose words are represented with a Markov model according to grammatical restrictions. Then a word classification information storage part 8, a word classifying process part 9, and a word group standard pattern storage part 10 are added. Then while individual word patterns are used for the process of a word series which appears at a high frequency by using the word group standard patterns that represent word groups obtained by grouping several words, the word group standard patterns are used for the process of a word series which is low in frequency, thereby recognizing diverse input sentences without relaxing the grammatical restrictions and increasing the throughput.
KUROIWA SHINGO
YAMAMOTO SEIICHI