SPEECH RECOGNITION DEVICE, SPEECH RECOGNITION METHOD, SPEECH RECOGNITION PROGRAM, AND INFORMATION RECORDING MEDIUM

摘要

PROBLEM TO BE SOLVED: To provide a speech recognition device etc., which eliminates speaker adaptation of an unnecessary word model to speech information corresponding to a speech and can improve the recognition rate of a keyword. SOLUTION: The speech recognition device which performs a keyword recognition is equipped with a database 11 which accumulates and stores keyword models for a plurality of keywords and also accumulates and stores unnecessary word models for a plurality of unnecessary words, a speaker adaptation processing part 9 which updates the respective models through speaker adaptation processing, and a recognition result decision part 8 which collates speech information with the updated keyword models and unnecessary word models. The speaker adaptation processing part 9 updates the unnecessary word models by setting an unnecessary word change vector indicating the change between unnecessary word models which are not updated and updated smaller than a corresponding unnecessary word change vector when the same update processing as update processing of a keyword model which is recognized in the past is applied to update of the unnecessary word model according to the keyword. COPYRIGHT: (C)2005,JPO&NCIPI