摘要 |
<p><P>PROBLEM TO BE SOLVED: To provide a speaker specification device in which a recognition rate hardly falls for a long period of time while the amount of calculation and memory capacity, etc., necessary for speaker specification processing are suppressed to a lowest possible rate. <P>SOLUTION: The speaker specification device for comparing a speaker model indicating the acoustic feature of utterance by the speaker and a feature quantity signal corresponding to the utterance of the speaker and specifying the corresponding speaker comprises previously memorizing a preliminary speaker model RSV corresponding to the intrinsic speaker and higher in the recognition rate of the speaker than the speaker model RM in addition to the registered speaker model RM used for the intrinsic speaker specification (step S11), and replacing the registered speaker model RM with the preliminary speaker model RSV when the recognition rate in the registered speaker model RM is lowered (step S15). <P>COPYRIGHT: (C)2009,JPO&INPIT</p> |