To reduce the amount of storage of voice recognition resource, to prevent the occurrence of misconception of numerical values of approximate voice and to improve the recognition rate of numerical values by generating the voice recognition resource and limiting the number of figures that are voice recognized based on the numerical values stored in a database.

The device is provided with a voice input section 17 which inputs voice of a speaker, a data table 35a which stores plural numerical values (merchandise unit prices), a voice recognition resource generating section 34 which extracts only the number used in every figure of the prices stored in the table 35a and sets the extracted number as the number to be recognized for every figure, and a voice recognition means 32 which extracts and recognizes the number for every figure of the numerical values that are voice inputted among the number of the corresponding figure of the voice recognition resource generated by the section 34.