Full-audio Detection on SONICS full audio
99.99RecallFusion Segment Transformer
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Fusion Segment TransformerFeature Extractor=MERT2026.01 | 99.99 | 99.99 | 100 | 99.99 | 99.99 | 100 | |
| MERT (Standalone)Architecture=Baseline (Linear Head)2026.01 | 99.96 | 99.96 | 100 | 99.96 | 100 | 99.96 | |
| Segment TransformerFeature Extractor=MERT2026.01 | 99.96 | 99.92 | 99.88 | 99.92 | 99.99 | 99.87 | |
| Fusion Segment TransformerFeature Extractor=Music2vec2026.01 | 99.88 | 99.94 | 99.99 | 99.93 | 100 | 99.99 | |
| Fusion Segment TransformerFeature Extractor=FXencoder2026.01 | 99.64 | 99.74 | 99.81 | 99.73 | 99.98 | 99.83 | |
| Segment TransformerFeature Extractor=FXencoder2026.01 | 99.61 | 99.55 | 99.53 | 99.57 | 99.95 | 99.48 | |
| Fusion Segment TransformerFeature Extractor=Muffin2026.01 | 97.75 | 98.28 | 98.63 | 98.19 | 99.78 | 98.76 | |
| SpecTTTra-αArchitecture=Baseline2026.01 | 96 | — | — | 97 | — | 99 | |
| Segment TransformerFeature Extractor=Music2vec2026.01 | 95.46 | 95.55 | 96.09 | 95.77 | 99.07 | 95.66 | |
| ConvNeXtArchitecture=Baseline2026.01 | 95 | — | — | 96 | — | 98 | |
| EfficientViTArchitecture=Baseline2026.01 | 94 | — | — | 95 | — | 97 | |
| Segment TransformerFeature Extractor=Wav2vec 2.02026.01 | 87.26 | 74.73 | 71.27 | 78.46 | 74.12 | 60.75 | |
| SpecTTTra-βArchitecture=Baseline2026.01 | 86 | — | — | 92 | — | 99 | |
| ViTArchitecture=Baseline2026.01 | 82 | — | — | 89 | — | 98 | |
| SpecTTTra-γArchitecture=Baseline2026.01 | 79 | — | — | 88 | — | 99 | |
| Fusion Segment TransformerFeature Extractor=Wav2vec 2.02026.01 | 75.68 | 77.58 | 77.04 | 76.35 | 85.63 | 79.33 |