PARAMETER ESTIMATION METHOD BY MIXED GAUSSIAN DISTRIBUTION, PATTERN RECOGNITION METHOD AND VOICE RECOGNITION METHOD USING SAME, APPARATUS USING THOSE METHODS, PROGRAM, AND RECORDING MEDIUM RECORDED WITH PROGRAM

摘要

PROBLEM TO BE SOLVED: To provide a method which can obtain Gaussian distributions containing important components with less parameters through a small amount of calculation for each Gaussian distribution and realize more likely parameter estimation for the unknown data, a pattern recognition method and a voice recognition method using this method, a system using those methods, and a program and a recording medium recorded with this program.;SOLUTION: The parameters of the all covariant mixed Gaussian distributions are estimated to obtain the proper value for the all covariant matrices of each Gaussian distribution. Selecting the minimum q which makes the rate of the total sum of the proper values other than the upper q proper values in the obtained proper values less than the predetermined threshold, a constraint covariant matrix is obtained to use as the parameters of the Gaussian distribution. This procedure is repeated for each Gaussian distribution.;COPYRIGHT: (C)2006,JPO&NCIPI