APPARATUS AND METHOD FOR LEARNING HMM, COMPUTER PROGRAM, AND COMPUTER

摘要

PROBLEM TO BE SOLVED: To enable high-precision automatic phoneme segmentation by using HMMs. SOLUTION: A learning apparatus 20 for HMM includes a learning data selection part 32 and an HMM selection part 34 which supply learning data with a demarcated phoneme boundary to an HMM 36, a loss function calculation part 38 which calculates a value of a loss function according to a likelihood that the HMM outputs and the phoneme boundary of the learning data, and an optimization part 22 which optimizes the HMM by controlling the learning data selection part 32, HMM selection part 34, and loss function calculation part 38 according to identification learning algorithm so that the calculated value of the loss function becomes minimum. COPYRIGHT: (C)2005,JPO&NCIPI