To make automatically a speech recognition dictionary constructible by using information on the web and to make an input speech recognizable with high precision.
The speech recognition dictionary is created by using log information of a search engine, etc., to recognize the input speech and the search engine is requested to make a search based upon the specified recognition result for obtaining the number of search items as to the recognition result; when the obtained number of search items is larger than the certain number of items, dictionary data of the recognition results is deleted from the created speech recognition dictionary to automatically structure the speech recognition dictionary. Then the input speech is recognized, a plurality of recognition results candidates are selected according to the likelihood and specified to request a search to the search engine, and the number of search items as to those recognition result candidates is obtained to determine the final recognition result of the input speech, thereby realizing high-precision recognition of the input speech.
OMORI NOBUYUKI
MIYAMOTO MASARU
INAGAKI HIROTO
