A method for generating voice dictionary is disclosed which makes it possible to improve the accuracy of speaker identification. A method according to an aspect of the present disclosure includes: acquiring voices of a plurality of unspecified speakers; acquiring noise in a predetermined place; superimposing the noise onto the voices of the plurality of unspecified speakers; and generating, on the basis of the features of the voices of the plurality of unspecified speakers, unspecified speaker voice dictionary that is used for generating personal voice dictionary for identifying a target speaker.