From: Three-stage training and orthogonality regularization for spoken language recognition
ID | Model | \(\lambda\) | channel-test | noisy-test | ||
---|---|---|---|---|---|---|
 |  |  | Cavg% | EER% | Cavg% | EER% |
0 | Top 1 on the OLR20 Challenge [51] | - | 2.39 | 2.47 | 3.47 | 4.07 |
1 | Top 2 on the OLR20 Challenge [51] | - | 4.21 | 4.51 | 4.76 | 4.87 |
2 | Top 3 on the OLR20 Challenge [51] | - | 4.77 | 4.82 | 5.38 | 5.60 |
3 | The best two-stage model | 0.1 | 3.90 | 7.25 | 2.06 | 4.20 |
4 | The best three-stage model (ours) | 0.1 | 1.88 | 3.57 | 1.86 | 3.74 |