| 摘要 |
<p>PROBLEM TO BE SOLVED: To provide a speaker matching device which facilitates registration, reduces storage capacity, and is robust against mispresentation by sound recording by combining speaker matching based upon a registered word standard pattern of the very person and speaking content matching based upon the voice of an unspecified speaker with each other. SOLUTION: An indication part 102 shows a specific text representing speaking contents which sound be spoken together with a registered word to a speaker by using sound, video, etc. Once the specific text and a speakers voice of the registered word are inputted, a spoken text confirmation part 107 confirms that the speaking contents of the speech pattern made to correspond to a standard pattern receiving the specific text is the same as the specific text. A similarity calculation part 111 finds the similarity between the speaker standard pattern corresponding to an inputted ID and the voice pattern made to correspond to the registered word. A decision part 112 judges that the speaker is a speaker that the speaker ID indicates on condition that the spoken text confirmation part 107 confirms that the specific text is spoken and the similarity found by the similarity calculation part 111 is larger than a predetermined threshold value.</p> |
