PAD system based on MFCC features and GMM-based classifier. Presentation attacks from AVspoof.
dev_eer | 0.0566202 |
---|---|
dev_eer_threshold | 0.552121 |
dev_far | 0.0565837 |
dev_frr | 0.0566567 |
test_far | 0.0561404 |
test_frr | 0.0814204 |
test_hter | 0.0687804 |
dev_negatives | 38580 |
dev_positives | 4995 |
test_negatives | 43320 |
test_positives | 5576 |
dev_scores | |
test_scores | |
dev_roc | |
test_roc |
19 MFCCs with deltas and double-deltas were used as features. Two GMMs are built for real data and spoof attacks and the evaluation score is computed as a log-likelihood between these two GMMs.
Updated | Name | Actions | |
---|---|---|---|
July 15, 2016 | pkorshunov/MFCC-GMM-based-PAD-system (Performance of MFCC-GMM-based PAD system on presentation attacks) | ||
July 15, 2016 | pkorshunov/Compare-PAD-systems (Performance of MFCC-GMM-based and LBP-LR-based PAD systems) |