摘要 |
PROBLEM TO BE SOLVED: To reduce the processing quantity needed for search processing for voice recognition by decreasing paths which should be searched for when an optimum path is searched for by a viterbi search. SOLUTION: This method is the voice recognizing method of a voice recognizing device which inputs a voice signal and recognizes a voice; and the inputted voice signal is converted from analog into digital and based on the digital signal, a feature vector series is extracted (S12). Then a search space which allows the transition of a specific feature vector series between HMMs is prescribed as a search space prescribed by the feature vector series and HMMs 16 prepared by voice units and in the space, the optimum path having the maximum sound likelihood to the voice signal is searched for to find a recognition result (S14), so that the recognition result is outputted (S15).
|