Table 5 Experimental results on Task 1A. Mixup Normalization and event extraction block are efficient on unseen devices (S4–S6) on TAU Urban AcousticScenes 2020 Mobile, Development dataset
From: Deep semantic learning for acoustic scene classification
| Â | Devices | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
Method | A | B | C | S1 | S2 | S3 | S4 | S5 | S6 | Overall |
Mini-SegNet+BN | 74.24 | 71.43 | 74.77 | 63.93 | 66.36 | 66.66 | 63.33 | 63.63 | 55.45 | 65.93\(\pm {0.7}\) |
Mini-SegNet+IN | 78.48 | 72.64 | 71.73 | 70.30 | 67.27 | 73.94 | 67.27 | 65.15 | 65.45 | 69.82\(\pm {0.4}\) |
Mini-SegNet +MixupNorm | 78.27 | 72.82 | 72.03 | 68.45 | 68.18 | 74.24 | 68.45 | 66.52 | 66.36 | 70.11\(\pm {0.3}\) |
Mini-SegNet +MixupNorm +EventExtractionBlock | 76.67 | 71.34 | 74.47 | 70.61 | 69.39 | 71.52 | 69.70 | 69.70 | 67.85 | 70.97\(\pm {0.3}\) |