| 摘要 |
<p>PROBLEM TO BE SOLVED: To provide a device and a method for voice authentication and a program which can accurately extract a voiced part useful for authentication to improve the authentication rate. SOLUTION: A feature extraction processing part 11b calculates cepstrum with fine precision as to a voice section of an input voice, extracts a voice part (voiced sound) according to a peak value appearing its high quefrency part, and calculates a slightly rough melcepstrum as to the extracted voice part, extracts its low quefrency part as feature data, and stores them in a registered data buffer 11c. When a new voice is inputted, a feature extraction processing part 11b extracts new feature data as to the voice part of the input voice and a feature matching part 11d matches the new feature data against the registered feature data by DP matching to authenticate the new voice as the registered voice according to the matching result.</p> |
