A speaker recognition method for an integrated voice recognition and speaker recognition system is provided to maximize the personality of information by speaker by using a feature parameter weight method or minimize a vector quantization error by using a multi-codebook searching method while increasing user security by integrating a voice recognition system and a speaker recognition system. A speaker recognition method for an integrated voice recognition and speaker recognition system comprises the following steps of: performing status segmentation by using a back-propagation path obtained in Viterbe operation for calculating a pre-threshold value after generating a personal model in the speaker recognition system(S61); obtaining a personal weight according to status by calculating the intra-speaker variance according to difference in number regarding an MFCC(Mel-scaled Frequency Cepstral Coefficient) according to status and an Euclid distance between world models obtained through the status segmentation(S62); and creating an observation sequence by performing multi-vector quantization by using the personal weight by status and performing the speaker recognition through the observation sequence(S63,S64).
展开▼