4-class respiratory sound classification on ICBHI 60-40% split official (test)
89.49SpecificityToikkanen et al. (BTS++[k = 30])
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Toikkanen et al. (BTS++[k = 30])Backbone=CLAP, Pretraining Data=LA, Venue=INTERSPEECH‘252026.06 | 89.49 | 41.89 | 65.69 | |
| Late-Fusionarchitecture=Inc-03 + VGG14, pretrain=IN2023.05 | 85.6 | 30 | 57.3 | |
| Jeong et al. (PAFA)Backbone=BEATs, Pretraining Data=AS, Venue=INTERSPEECH‘252026.06 | 82.05 | 47.63 | 64.84 | |
| Bae et al. (Patch-Mix CL)architecture=AST, pretrain=IN + AS2023.12 | 81.66 | 43.07 | 62.37 | |
| Patch-Mix CL (ours)architecture=AST, pretrain=IN + AS2023.05 | 81.66 | 43.07 | 62.37 | |
| Bae et al. (Patch-Mix CL)Backbone=AST, Pretraining Data=IN + AS, Venue=INTERSPEECH‘232026.06 | 81.66 | 43.07 | 62.37 | |
| Kim et al. (BTS)Backbone=CLAP, Pretraining Data=LA, Venue=INTERSPEECH‘242026.06 | 81.4 | 45.67 | 63.54 | |
| SE+SAarchitecture=ResNet182023.05 | 81.25 | 17.84 | 49.55 | |
| Xiao et al. (LungAdapter)Backbone=AST, Pretraining Data=IN + AS, Venue=INTERSPEECH‘242026.06 | 80.43 | 44.37 | 62.4 | |
| SG-SCLarchitecture=AST, pretrain=IN + AS2023.12 | 79.87 | 43.55 | 61.71 | |
| Kim et al. (SG-SCL)Backbone=AST, Pretraining Data=IN + AS, Venue=ICASSP‘242026.06 | 79.87 | 43.55 | 61.71 | |
| DASS (Lung-SRAD)Backbone=DASS, Pretraining Data=AS, Venue=-2026.06 | 79.53 | 49.42 | 64.48 | |
| Nguyen et al. (CoTuning)architecture=ResNet50, pretrain=IN2023.12 | 79.34 | 37.24 | 58.29 | |
| Nguyen et al. (CoTuning)architecture=ResNet50, pretrain=IN2023.05 | 79.34 | 37.24 | 58.29 | |
| Nguyen et al. (StochNorm)architecture=ResNet50, pretrain=IN2023.05 | 78.86 | 36.4 | 57.63 | |
| Bae et al. (Fine-tuning)architecture=AST, pretrain=IN + AS2023.12 | 77.14 | 41.97 | 59.55 | |
| Fine-tuning (ours)architecture=AST, pretrain=IN + AS2023.05 | 77.14 | 41.97 | 59.55 | |
| Bae et al. (Fine-tuning)Backbone=AST, Pretraining Data=IN + AS, Venue=INTERSPEECH‘232026.06 | 77.14 | 41.97 | 59.55 | |
| DATarchitecture=AST, pretrain=IN + AS2023.12 | 77.11 | 42.5 | 59.81 | |
| DASS (Spectral-Aware Regularization)Backbone=DASS, Pretraining Data=AS, Venue=-2026.06 | 76.72 | 47.72 | 62.22 | |
| Moummad et al. (SCL)architecture=CNN6, pretrain=AS2023.12 | 75.95 | 39.15 | 57.55 | |
| Moummad et al. (SCL)architecture=CNN6, pretrain=AS2023.05 | 75.95 | 39.15 | 57.55 | |
| DASS (Fine-tuning)Backbone=DASS, Pretraining Data=AS, Venue=-2026.06 | 74.68 | 47.43 | 61.06 | |
| Ren et al.architecture=CNN8-Pt2023.12 | 72.96 | 27.78 | 50.37 | |
| Ren et al.architecture=CNN8-Pt2023.05 | 72.96 | 27.78 | 50.37 | |
| CNN-MoEarchitecture=C-DNN2023.12 | 72.4 | 21.5 | 47 | |
| CNN-MOEarchitecture=C-DNN2023.05 | 72.4 | 21.5 | 47 | |
| RespireNet (CBA+BRC+FT)architecture=ResNet34, pretrain=IN2023.12 | 72.3 | 40.1 | 56.2 | |
| RespireNet (CBA+BRC+FT)architecture=ResNet34, pretrain=IN2023.05 | 72.3 | 40.1 | 56.2 | |
| LungAttnarchitecture=ResNet-Att2023.05 | 71.44 | 36.36 | 53.9 | |
| RespireNetarchitecture=ResNet34, pretrain=IN2023.05 | 71.4 | 39 | 55.2 | |
| Wang et al. (Splice)architecture=ResNeSt, pretrain=IN2023.12 | 70.4 | 40.2 | 55.3 | |
| Wang et al. (Splice)architecture=ResNeSt, pretrain=IN2023.05 | 70.4 | 40.2 | 55.3 | |
| Moummad et al.architecture=CNN6, pretrain=AS2023.05 | 70.09 | 40.39 | 55.24 | |
| Chang et al. (CNN8-dilated)architecture=CNN8-dilated2023.12 | 69.92 | 35.85 | 52.89 | |
| Chang et al. (CNN8-dilated)architecture=CNN8-dilated2023.05 | 69.92 | 35.85 | 52.89 | |
| LungBRNarchitecture=bi-ResNet2023.05 | 69.2 | 31.12 | 50.16 | |
| ARSC-Netarchitecture=bi-ResNet-Att2023.05 | 67.13 | 46.38 | 56.76 | |
| LungRN+NLarchitecture=ResNet-NL2023.05 | 63.2 | 41.32 | 52.26 | |
| Chang et al. (ResNet-dilated)architecture=ResNet-dilated2023.12 | 50.22 | 51.83 | 51.02 | |
| Chang et al. (ResNet-dilated)architecture=ResNet-dilated2023.05 | 50.22 | 51.83 | 51.02 |