| 摘要 |
<P>PROBLEM TO BE SOLVED: To obtain a highly accurate recognition result in accordance with a topic. <P>SOLUTION: A voice recognition apparatus includes: an acoustic analysis part 50 for calculating an acoustic feature value on the basis of voice data; a correct answer word search part 60 for finding out an acoustic score in each language expression on the basis of the acoustic feature value and an acoustic model corresponding to a pronunciation network, finding out a language score, searching a correct answer candidate word string on the basis of the acoustic score and the language score, and generating recognition result text information; a topic tracking part 80 for extracting a speech correspondence text corresponding to the recognition result text from topic information; a language model adaptation part 90 for extracting a speech correspondence vicinity text including the speech correspondence text from the topic information, acquiring related text information related to the speech correspondence vicinity text from a text information source 2, and adapting and updating a language model on the basis of the related text information; and an updating part 62 which, when the language model is updated, updates the speech network and the language score memory on the basis of the adaptive language model. <P>COPYRIGHT: (C)2013,JPO&INPIT |
