PURPOSE: To recognize a desired word by providing a voice feature extracting section, temporary feature value memory, a pattern identifying section, a threshold setting section, a voice recognition control means and a means for deleting an additional and temporary pattern so that a threshold value is temporarily updated upon rejection.

CONSTITUTION: A voice from a microphone 1 is patternized by a sound feature extracting section 2 and is then stored in a temporarily feature value memory 3. Then, it is identified by a pattern identifying section 5 so as to determine a first candidate. A result determining section 6 delivers the first candidate as a result of recognition if the resemblance of the first candidate is higher than a threshold value set by a threshold setting section 7, but if it is lower, the determining section delivers no output so that a temporary pattern of the voice in the memory 3 is temporarily and additionally created in standard pattern memory 4. The same sound is again issued, the threshold value is temporarily decreased, and if a highest resemblance in a standard pattern exceeds this threshold value, the determining section 6 delivers the 'additional and temporary pattern' in the memory 4, as a result of recognition. When a limited number of seconds set on a timer 9 is exceeded, the threshold value is returned to its original value. Further, if no next voice is issued within a predetermined time, a means for deleting the additional and temporary pattern deletes the additional and temporary pattern in the memory 4.