| 摘要 |
PROBLEM TO BE SOLVED: To adapt a language model to dialogs and perform speech recognition with a high recognition rate without the need to learn a large amount of data, without creating evaluation data, without a large amount of preparations or calculations. SOLUTION: The speech recognition technique includes the steps of: recognizing dialog speech; extracting feature value from an speech signal; performing speech recognition using the feature value obtained from the speech signal containing the utterance contents of a predetermined speaker A, a sound model and the language model before adaption to determine a recognition result A'; determining the language model after adaption using the recognition result A' only and the language model before adaption; and performing speech recognition using the feature value obtained from the speech signal containing the utterance contents of a speaker B other than the predetermined speaker, the sound model and the language model before adaption to determine a recognition result B'. COPYRIGHT: (C)2011,JPO&INPIT
|
