SEMI-SUPERVISED SIGNAL RECOGNITION RETRIEVAL DEVICE, SEMI-SUPERVISED SIGNAL RECOGNITION RETRIEVAL METHOD AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To improve the accuracy of a signal recognition and signal retrieval by accurately learning the relationship between a signal and language information even when only a small amount of the signal associated with additional information (language information) can be used.

SOLUTION: A semi-supervised signal recognition retrieval device 10 includes: an accumulation signal feature extraction part 1 which extracts an accumulation signal feature from an accumulation signal; an accumulation additional information feature extraction part 2 which extracts an accumulation additional information feature from accumulation additional information; an accumulation latent variable extraction part 3 which extracts an accumulation latent variable from an accumulation signal feature aggregation and an accumulation additional information feature aggregation; a topic model learning part 4 which extracts a topic model from the accumulation signal feature aggregation, the accumulation additional information feature aggregation, and an accumulation latent variable aggregation; an input additional information feature extraction part 5 which extracts an input additional information feature when input additional information is inputted; and a signal retrieval part 6 which gives the input additional information feature to the topic model and selects a signal related to the input additional information.

COPYRIGHT: (C)2012,JPO&INPIT