摘要 |
PURPOSE:To make it possible to fulfill the accurate recognizing collation of an input voice by giving a large weight to an uneven sampling point in a period when sampling points are coarse. CONSTITUTION:Feature-quantity time-series information which corresponds to an unknown input voice is set to set part 11, and weight time-series information is to set part 10; and reference feature quantity time-series information of each registered word is stored in registration part 9, and weight time-series information is in registration part 8. In this state, reference feature-quantity time-series information and weight time-series information are supplied, registered word by registered word, to DP processor 12 under the control of control part 15 so as to attain sequential collation by processor 12, so that the most matching one will be obtained through decision 13 before being outputted 14. Consequently, the recognizing collation of the input voice can be performed accurately by giving a large weight to an uneven sampling point in the period when sampling points are coarse. |