Disclosed is a voice registration method for voice recognition, comprising the steps of analyzing a spectrum of a sound signal inputted from the outside; extracting predetermined language units for a speaker recognition from a voice signal in the sound signal; measuring the loudness of each language unit; collecting voice data on registered(background) speakers including loudness data of the plurality of background speakers as a reference onto voice database; determining whether the loudness of each language unit is within a predetermined loudness range based on the voice data base; learning each language unit by using a multi-layer perceptron in the case that at least a predetermined number of language units arc within the predetermined loudness range; and storing data on the learned language unit as data for recognizing the speaker. With this configuration, loudness of a speaker is considered at learning for registering his/her voice and at verifying a speaker, so that it is possible to more correctly verify the speaker.