ADAPTIVE MODEL LEARNING METHOD AND ITS DEVICE, SOUND MODEL CREATING METHOD FOR SPEECH RECOGNITION USING THE SAME AND ITS DEVICE, SPEECH RECOGNITION METHOD USING THE SOUND MODEL AND ITS DEVICE, PROGRAMS FOR THE DEVICES, AND RECORDING MEDIUM OF THE PROGRAMS

摘要

PROBLEM TO BE SOLVED: To provide a new adaptive model learning method and its device etc. in which concept of coarse/fine learning is introduced into a conventional estimation method of an affine transform parameter. SOLUTION: The adaptive model learning device comprises: an initial model storage section and a feature amount vector conversion section; a coarse/fine class setting section; an affine transform parameter learning section; and a model parameter conversion section. The coarse/fine class setting section decomposes each Gaussian distribution of a mixed Gaussian distribution into a mean vector and a linear transformation matrix, and generates the affine transform parameter with coarse/fine expression, and a generalized homogeneous vector composed of a partial vector and a mean vector parameter. The affine transform parameter leaning section estimates an estimation value of the affine transform parameter from an initial model, a feature amount vector, the affine transform parameter, and the generalized homogeneous vector, by a statistic learning method for a potential model. COPYRIGHT: (C)2008,JPO&INPIT