PURPOSE: To speed up processing by classifying voices of plural persons as well as a single person and setting the analysis sections of cepstrum peaks of respective groups at the time of registration.

CONSTITUTION: An analysis section setting section 4 calculates the optimum analysis sections corresponding to the peaks of the cepstrum supplied from a peak detecting section 2 and supplies the same to an analysis section classifying section 5. This setting section also supplies the analysis section data supplied from an analysis section memory 6 or the predetermined analysis section data to the peak detecting section 2 by the instruction of the analysis section classifying section. The analysis section classifying section 5 compares the optimum analysis section data and the analysis section data stored in the analysis section memory 6, classifies and processes the data and controls the data memory to the analysis section memory 6. The analysis sections made with the voices of the plural persons are classified and set by groups or independently at the time of registration and, therefore, the setting by limiting the analysis sections of the peak detection at the time of recognition is possible. The processing of the voice discrimination is speeded up in this way.