PURPOSE: To prevent a recognizing mistake by resetting the digital voice data corresponding to the identification code of the digital sound data discriminated from a new voice input to the sound data which is discriminated this time after reading said identification code out of a voice memory which stores with sorting the digital voice data and identification codes.

CONSTITUTION: The voice input given from a mike 7 is transferred to a register A in a CPU17 via a buffer VB. At the same time, the digital voice data M(1,1)... which are preset in three types for each department are read out to a register B. Then an error is calculated by an arithmetic circuit 21 and written to an auxiliary memory 23. When a voice searching action is over, the sum total m(1,1)+ m(1,2)+... of errors is calculated. The digital sound data of this time which is set to the buffer VB is written and reset to the memory area of the digital sound data having the minimum error among three digital voice data on the department having the minimum error.