PROBLEM TO BE SOLVED: To provide a device for creating a language model capable of creating a language model which improves voice recognition accuracy even when voice recognition accuracy of an uttered voice of a speaker responding to utterance of a speaker with high voice recognition accuracy is low.;SOLUTION: A device for creating a language model comprises: utterance characteristic extraction means 101 for extracting an utterance characteristic which is a characteristic of utterance of a speaker with high voice recognition accuracy participating in a conversation; language model study data selection means 102 for selecting study data for language model creation by using the utterance characteristic extracted by the utterance characteristic extraction means 101, and a conversation corpus that associates a written text of an uttered voice in a specific conversation, a voice characteristic amount obtained from the uttered voice or conversation, and speaker information obtained from the uttered voice or conversation; and language model creation means 103 for creating the language model by preferentially using the study data selected by the language model study data selection means 102.;COPYRIGHT: (C)2013,JPO&INPIT
展开▼
