A dynamic speech recognition dictionary generating unit extracts, from phrases stored in a speech recognition dictionary, phrases of which heads match a head of a character string inputted through a character string input unit and generates a dynamic speech recognition dictionary that stores difference phrases that are each part of an extracted phrase excluding a common phrase that is common among the extracted phrases. A speech recognition unit carries out recognition of a user utterance by using the dynamic speech recognition dictionary. An input character string confirming unit confirms an input character string candidate that includes a recognized difference phrase as an input character string.