PURPOSE: To prevent a decline in recognition of the title device by simulating a real time sequential correlation factor by adding the time sequential correlating factors of the 1st and 2nd correlation memories to each other and using the simulated factor as a template when an additivity is established between voices and background noises and the background noises fluctuate with a time constant larger than the length of a word to be recognized.

CONSTITUTION: When voices and background noises statistically independent and, at the same time, an additivity is established between the voices and noises in the dimension of correlating factors and the background noises fluctuate at a time constant longer than the length of a word to be recognized, the real time sequential correlating factor at the moment when voices to be recognized are inputted is simulated by adding time sequential correlating factors stored in the 1st and 2nd correlation memories 2 and 3 to each other and the simulated factor is used as a template. Therefore, the background noise conditions when the template is prepared and background noise conditions when voices 4to be recognized are inputted can be approximated to each other and lowering of the recognition rate can be prevented.