PURPOSE: To obtain a stable update accelerating function by judging whether a standard pattern is updated or not quantitatively from misrecognition information obtained from many input data.

CONSTITUTION: An inputted speech is analyzed by a recognition processing part 2 and converted into a series of feature parameters and held as an input pattern, and this input pattern is matched to previously generated standard patterns in a standard pattern storage part 3 to obtain a recognition result. This recognition result while sent to a recognition result processing part 4 is displayed on a display 5 together with other recognition candidates. A user selects and corrects the recognition result over a look at the display 5, and the result is displayed on the display 5 and also stored as update acceleration information on the standard pattern; when certain conditions are met, a display promoting the update of the standard pattern appears on the display 5. Consequently, the objective and quantitative update accelerating function is obtained.