Urban Sound Classification on UrbanSound 8k
87.2AccuracyGated Trans.
Evaluation results
| Method | Links | |
|---|---|---|
| Gated Trans.Encoder Type=Fusion2026.06 | 87.2 | |
| WQ-FusionEncoder Type=Fusion2026.06 | 87.1 | |
| Adapt. and Trans.Encoder Type=Fusion2026.06 | 86.9 | |
| BEATs (300M) BalancedParameter Count=300M, Pre-training Mixture=Balanced (40:30:30)2026.01 | 86.3 | |
| Ensemble (Submission)Ensemble Strategy=Embedding Concatenation2026.01 | 86.2 | |
| BEATs (300M) SpeechParameter Count=300M, Pre-training Mixture=Speech-heavy (70:15:15)2026.01 | 85.7 | |
| Concat.Encoder Type=Fusion2026.06 | 85.7 | |
| AudioMAEEncoder Type=Single2026.06 | 85.4 | |
| BEATs (90M) iter3Parameter Count=90M, Iteration=32026.01 | 85.3 | |
| Qwen2-Audio-7BEncoder Type=Single2026.06 | 84.7 | |
| Dasheng 1.2BParameter Count=1.2B2026.01 | 84.6 | |
| Challenge BaselineDescription=Best among Dasheng-base, data2vec, and Whisper2026.01 | 83.5 | |
| Whisper-LargeEncoder Type=Single2026.06 | 83.4 | |
| Dasheng-BaseEncoder Type=Single2026.06 | 82.9 | |
| F3-TokenizerProbing=Frozen representation2026.06 | 74.22 | |
| Ming-UProbing=Frozen representation2026.06 | 72.87 | |
| w/o LLMProbing=Frozen representation2026.06 | 70.6 | |
| WhisperProbing=Frozen representation2026.06 | 68.7 | |
| w/o RQProbing=Frozen representation2026.06 | 62.8 | |
| SEPTEvaluation Role=Target2026.06 | 59.34 | |
| DePTEvaluation Role=Target2026.06 | 56.25 | |
| CoCoOpEvaluation Role=Target2026.06 | 55.18 | |
| Zero-shotEvaluation Role=Target2026.06 | 52.57 | |
| SubT†Evaluation Role=Target2026.06 | 50.98 | |
| CoOpEvaluation Role=Target2026.06 | 50.82 | |
| CLIP-AdapterEvaluation Role=Target2026.06 | 49.65 | |
| SubTEvaluation Role=Target2026.06 | 48.72 | |
| No repr.Probing=Frozen representation2026.06 | 28.5 | |
| KgCoOpEvaluation Role=Target2026.06 | 21.03 |