From: Comparison of semi-supervised deep learning algorithms for audio classification
Dataset | ESC-10 | UBS8K | GSC | |||
---|---|---|---|---|---|---|
Labeled fraction | 10% | 100% | 10% | 100% | 10% | 100% |
CNN models (literature) | - | 3.00 [38] | - | 14.50 [39] | - | 3.00 [40] |
Supervised | 32.00 ±6.17 | 8.00 ±5.06 | 33.80 ±4.82 | 23.29 ±5.80 | 10.01 | 4.94 |
+mixup | 36.00 ±5.22 | 8.33 ±4.56 | 31.41 ±5.56 | 22.04 ±5.99 | 8.83 | 3.86 |
+weak | 22.67 ±3.46 | 4.67 ±3.43 | 27.08 ±4.58 | 20.09 ±5.50 | 7.62 | 3.90 |
+weak+mixup | 24.67 ±4.92 | 4.67 ±1.39 | 23.75 ±4.73 | 17.96 ±3.64 | 6.58 | 3.00 |
+strong | 23.00 ±5.19 | 5.00 ±2.64 | 25.58 ±4.15 | 20.69 ±4.92 | 7.60 | 3.27 |
+strong+mixup | 24.00 ±8.71 | 5.00 ±4.25 | 24.73 ±4.42 | 18.52 ±4.38 | 6.86 | 2.98 |