JPH05232983

PURPOSE:To shorten a processing time and to prevent a recognition ratio from being dropped by reducing the volume of calculation in the voice recognition of an unspecified speaker independently of sexes and ages. CONSTITUTION:Respective frequency level-divided reference pattern groups 1101 to 1104 which are independently set up in each average frequency are stored in a storage part 1005, a previously determined vocabulary is inputted before the execution of voice recognition, the average frequency of the voice data is measured and judged by a frequency detection part 1001, a feature matching part 1006 is connected to the most matched reference pattern group out of the groups 1101 to 1104 through frequency level-divided reference pattern connecting circuits 1011 to 1013, and then voice recognition is executed. Thereby necessity for calculating the similarity of all reference patterns stored in the storage part 1005 can be eliminated.