AURAL SIGNAL RETRIEVING METHOD, AURAL SIGNAL STORING METHOD FOR SPEECH RETRIEVAL, AURAL SIGNAL RETRIEVING DEVICE, PROGRAM THEREFOR AND RECORDING MEDIUM FOR THE PROGRAM

摘要

PROBLEM TO BE SOLVED: To make retrievable speech data similar to a retrieval key at a high speed by shortening the time of calculation of a similar distance by a simple matching between fixed-length data in retrieval of the speech data.;SOLUTION: A time window segmentation part 12 extracts partial speech time-series feature quantities which are different in length from an aural signal to be retrieved by using time windows of a plurality of kinds of lengths and a partial speech time- series feature quantity length fixation part 13 linearly expands or contracts partial speech time-series feature quantities to a specified reference time window length and stores them in a speech storage part 14. In retrieval, a retrieval key speech time-series feature quantity extraction part 18 extracts a retrieval key speech time-series feature quantity vector with the length of a reference time window from the input speech signal of the retrieval key and a feature quantity information comparison part 16 calculates the similarity distance between the extracted vector and a speech time-series feature quantity vector to be retrieved to decide a vector having higher similarity as a retrieval result.;COPYRIGHT: (C)2003,JPO