Optimization of detection systems using a detection error tradeoff analysis criterion
View Patent ↗In detection systems, such as speaker verification systems, for a given operating point range, with an associated detection “cost”, the detection cost is preferably reduced by essentially trading off the system error in the area of interest with areas essentially “outside” that interest. Among the advantages achieved thereby are higher optimization gain and better generalization. From a measurable Detection Error Tradeoff (DET) curve of the given detection system, a criterion is preferably derived, such that its minimization provably leads to detection cost reduction in the area of interest. The criterion allows for selective access to the slope and offset of the DET curve (a line in case of normally distributed detection scores, a curve approximated by mixture of Gaussians in case of other distributions). By modifying the slope of the DET curve, the behavior of the detection system is changed favorably with respect to the given area of interest.
1. A computer implemented method of facilitating verification in a detection system, said method comprising steps of:
accepting input data comprising true target data and non-target data;
evaluating the input speech data and deriving scores from the input speech data; and
evaluating a criterion associated with a detection error tradeoff curve relating to said scores using an evaluation loop in which said criterion is minimized through successively adapting the scores until said scores reach a predetermined quality wherein the reduction of both a False Alarm Rate error and a Miss Rate error are simultaneously achieved.
2. The method according to claim 1 , wherein the detection system is a speaker detection system used in speaker verification.
3. The method according to claim 1 , wherein said step of evaluating the scores comprises accepting a Gaussian score distribution associated with the scores.
4. The method according to claim 1 , wherein said step of evaluating the scores comprises accepting a Gaussian-Mixture distribution associated with the scores.
5. The method according to claim 1 , wherein said step of deriving scores comprises using model parameters in deriving scores from the input data.
6. The method according to claim 5 , wherein said step of successively prompting revision of at least one aspect associated with the scores comprises successively prompting revision of the model parameters until the scores reach a predetermined quality.
7. The method according to claim 1 , further comprising:
providing a plurality of detection systems for evaluating the input data and deriving scores from the input data;
said step of deriving scores comprising using a predetermined weighting factor associated with each detection system in deriving scores from the input data.
8. The method according to claim 7 , wherein said step of successively prompting revision of at least one aspect associated with the scores comprises successively prompting revision of the weighting factors until a final score comprising a sum of weighted scores reaches a predetermined quality.
9. The method according to claim 1 , wherein the criterion associated with a detection error tradeoff is selectively applied to modify at least one of: the slope of a detection error tradeoff curve and the offset of a detection error tradeoff curve.
10. The method according to claim 9 , wherein the criterion associated with a detection error tradeoff is selectively applied to modify the slope and offset of a detection error tradeoff curve.
11. The method according to claim 10 , wherein the criterion associated with a detection error tradeoff is selectively applied to modify the slope and offset of a detection error tradeoff curve in a direction of interest and in a controlled fashion.
12. The method according to claim 9 , wherein the detection error tradeoff curve relates at least to a false alarm rate.
13. The method according to claim 9 , wherein the detection error tradeoff curve relates at least to a miss rate.