LEARNING DATA SELECTING DEVICE, LEARNING DATA SELECTING METHOD, PROGRAM AND RECORDING MEDIUM, AND ACOUSTIC MODEL GENERATING DEVICE, ACOUSTIC MODEL GENERATING METHOD, PROGRAM, AND RECORDING MEDIUM

摘要

PROBLEM TO BE SOLVED: To select a learning data useful for generating a highly accurate acoustic model for achieving high recognition performance in the learning data. SOLUTION: An adaptive acoustic model 151 which is adapted to a task is generated by learning a base acoustic model 141 with a task adaptive learning data 120 which are the learning data being adaptive to the task. Speech recognition is performed on the speech data of a learning data 200, and the recognition score (an adaptive recognition score) of speech recognition, when a grammar 161 obtained from the label of the learning data 200, and an adaptive acoustic model 151 are used, and the recognition score (a base recognition score) of speech recognition, when the grammar 161 and the base acoustic model 141 are used, are calculated. Among the learning data 200, the data passes the comparison evaluation of the adaptive recognition score and the base recognition score, are selected. COPYRIGHT: (C)2009,JPO&INPIT