PATTERN RECOGNIZING METHOD AND PATTERN RECOGNIZING DEVICE

摘要

PROBLEM TO BE SOLVED: To reduce a calculation quantity related to recognizing processing, and to improve recognizing accuracy by calculating a probability only on a characteristic vector belonging group of a recognizing input pattern judged by using a deciding wood. SOLUTION: The deciding wood making part 4 makes a deciding wood on the basis of HMM stored in the HMM storage part 2, and the deciding part storage part 5 stores the deciding wood being made. The leaf selecting part 7 selects a leaf being a tail end node by tracking the deciding wood according to a predetermined rule on the basis of a characteristic vector of a recognizing object delivered from the characteristic vector making part 1 at recognizing time and a node parameter stored in respective nodes of the deciding wood stored in the deciding wood storage part 5. The output probability calculating part 8 calculates an output probability by using a leaf parameter stored in a selected leaf and a discriminating function. The checkup part 9 makes a checkup calculation on the basis of the output probability to respective time series characteristic vectors made from a voice pattern of the recognizing object calculated by the output probability calculating part 8 to output a recognizing result of an input voice.