SPEAKER MODEL REGISTRATION DEVICE AND METHOD IN SPEAKER RECOGNITION SYSTEM AND COMPUTER PROGRAM

摘要

A speaker model registration device (10) in a speaker recognition system (1) includes: acquisition means (13) for acquiring n utterances (n is an integer not smaller than 2); similarity calculation means (15) for calculating a similarity indicating similarity between the n utterances acquired; and verification means (16) for verifying whether an inappropriate utterance having similarity not reaching a predetermined reference exists. The speaker model registration device (20) further includes: speaker model calculation means for calculating a speaker model according to the n utterances not including any inappropriate utterance if no inappropriate utterance exists as a result of the verification; and registration means (40) for registering the calculated speaker model as a speaker model for speaker recognition.