PURPOSE: To enable to recognize the object pattern of recognition at high accuracy even when the inputted object pattern of recognition belongs to a specific user or belongs to an unspecific user by inputting information on the supplier of the object pattern of recognition, and deciding whether the supplier is a specific user or unspecific user basing on the information.

CONSTITUTION: Continuous voice, which is the object pattern of recognition, inputted through a voice inputting section 1 is given to a acoustic analyzing section 2, and feature parameter of the inputted voice is formed. Time series of the feature parameter thus obtained is given respectively to a recognizing section 3 for specific speaker and a recognizing section 4 for unspecific speaker. The pattern of the inputted voice indicated by time series of the feature parameter and standard parameter are compared and collated, and the result of recognition, i.e. the result of collation, is outputted through a result of recognition outputting section 7. That is to say, the voice inputted as an object pattern of recognition is decided as to whether the voice is of a specific speaker or of an unspecific speaker according to speaker information, and recognized and processed flexibly and efficiently conforming to recognition mode set according to the result of decision, and the result of recognition is obtained at high accuracy and outputted.