| 摘要 |
<p><P>PROBLEM TO BE SOLVED: To provide a speaker specifying device capable of improving accuracy in specifying a speaker by appropriately updating a similarity threshold by flexibly matching it to progress of updating speaker feature information, while suppressing a calculation amount and a storage capacity etc. required for specifying the speaker to the minimum. <P>SOLUTION: The speaker specifying device includes: a speaker recognition section 5 in which a speaker model for indicating an acoustic feature of utterance by the speaker is compared with a feature signal So which corresponds to utterance of the speaker, and a similarity degree between them is detected, and in which the speaker is specified by comparing the detected similarity degree with the similarity threshold which is compared, when the similarity degree is detected; a parameter updating section 8 for updating the speaker model by using the feature signal So; and a threshold setting section 6 for updating the similarity threshold by using the detected similarity degree. <P>COPYRIGHT: (C)2009,JPO&INPIT</p> |
