| 摘要 |
<p>PROBLEM TO BE SOLVED: To lessen the user's burden at the time of threshold setting of speaker recognition by providing every user with thresholds TA and TB, setting the thresholds TA in such a manner that a feigning person acceptance rate is equaled to the feigning person acceptance rate of a recognition step and setting the threshold TB in such a manner that a person himself or herself abort rate is equaled to the person himself or herself abort rate of the recognition step. SOLUTION: As the first recognition step, the users are let to carry out utterance as a first theme of the light burden and every user is provided with the thresholds TA and TB (TA>TB). The threshold TA is so set that the feigning person acceptance rate when the first theme is carried out is equaled to the feigning person acceptance rate of the second recognition step and the threshold TB is so set that the person himself or herself abort rate is equaled to the person himself or herself abort rate in the second recognition step. A threshold setting section 15 sets the three thresholds T, TA and TB with each of the users' IDs in accordance with the characteristic data of the users stored in a characteristic quantity preservation section 13. The TA and TB are set dependently upon the person himself or herself abort rate, the feigning person acceptance rate and the decision rate in the theme 1.</p> |
