摘要 |
PROBLEM TO BE SOLVED: To provide a voice recognizing device which can recognize highly accurately a voice with less quantity of calculation. SOLUTION: In this device, a temporary pattern storage section 104 stores a partial standard pattern being the subject of calculation of the degree of similarity for an input pattern (by extracting feature quantity) to which a voice waveform from a voice analyzing section 102 is analyzed and the cumulative degree of similarity of these patterns out of standard patterns of recognition subject words of a standard pattern storage section 103, and a counting section 106 compares the cumulative degree of similarity with the present threshold value of a threshold value storage section 105 and counts the number of partial patterns having the large degree of similarity. A threshold counting section 107 counts a pruning threshold value in accordance with correspondence relation of the present threshold value and the numbers, a degree of similarity counting section 108 counts feature quantity and the degree of similarity of partial standard patterns having the large degree of similarity based on the pruning threshold value. A deciding section 109 decides a word corresponding to a standard pattern having the largest degree of similarity out of partial standard patterns as the recognized result. |