| 摘要 |
<p><P>PROBLEM TO BE SOLVED: To provide a sound model learning device capable of automatically creating a learning data of a sound model with high recognition rate, and creating the sound model applicable to various utterance styles, even when an off line subtitle in which a matching rate of the subtitle and an utterance content is low, is used. <P>SOLUTION: The sound model learning device comprises: a sound model storage section for storing a sound model in which a phoneme is related to a sound feature amount corresponding to the phoneme; a recognition processing section for performing voice recognition processing by using the sound model which is read from the sound model storage section, and outputting a recognition result data; a selection processing section for selecting a matching section on the basis of a text data corresponding to the voice, and of the recognition result data which are output by the recognition processing section, and outputting a set of the phoneme included in the matching section, and a sound feature amount corresponding to the phoneme, as the learning data; and a sound model adaptation section for updating the sound model by using the learning data which is output by the selection processing section. <P>COPYRIGHT: (C)2009,JPO&INPIT</p> |
