are not included in one of the utterances of the speaker who is speaking words, even if included in the speech of another speaker responding to the utterance, the other story improve the recognition accuracy of processing to recognize the speech of the person. Language model generation apparatus 18 for generating a language model used for speech recognition process includes an extraction unit 100 and the model generating unit 101. Extraction unit 100, based on the word or word sequence contained in the first speakers utterance of the first speakers and other speakers which enters into a dialogue to be recognized, the second data from data group 20 It is extracted. A data group 20, a first data is a word or word string included in the first speech, associated with the first data of the word or word string included in the second speech responsive to the first utterance a second data a related word is a data group of a collection of related words pairs associated to. Model generating unit 101, by using the second data extracted by the extraction unit 100, generates a language model corresponding to the speech of the other speaker.
展开▼
