VOICE RECOGNITION SYSTEM, VOICE RECOGNITION METHOD, AND VOICE RECOGNITION PROGRAM

摘要

It is possible to rapidly perform a recognition process even when much garbage is present in a grammar. Provided is a voice recognition system which includes a first voice recognition process unit and a second voice recognition process unit. The first voice recognition process unit subjects an input voice feature amount to a voice recognition process based on a first grammar and generates a recognition hypothesis graph indicating a structure of a hypothesis derived from the first grammar together with a score associated with respective connections of recognition units. The second voice recognition process unit subjects the input voice feature amount to a voice recognition process based on a second grammar defined so as to receive sections other than a keyword in the input voice as garbage sections and outputs a recognition result from the overall score of the hypothesis derived according to the second grammar. The second voice recognition process unit acquires the structure and the score of the garbage section from the recognition hypothesis graph.