To automatically switch domains to precisely recognize voices while suppressing increase in a calculation cost.
A voice recognition device realized by a program for operating a computer as a voice recognition device and a sentence classification device uses hierarchized language models and comprises; a general-purpose voice recognition part 220 which uses general-purpose language models in higher hierarchies to perform voice recognition; a topic detection part 222 which detects topics of the recognition result by calculating distances between the recognition result and topic-classified language models in lower hierarchies; a topic storage part 230 for storing selected topics; a topic-classified voice recognition part 226 which uses language models corresponding to stored topics to perform voice recognition; and a selection part 236 which selects one of the recognition result of the general-purpose voice recognition part 220 and that of the topic-classified voice recognition part 226, which has higher reliability.
KAWAHARA TATSUYA
MATSUI TOMOKO
NAKAMURA SATORU
Next Patent: METHOD OF MANUFACTURING POSITIVE CHARGE TYPE TONER