Sound Classification on FSD Kaggle 18
90.5ScoreUniWhisper
Evaluation Results
| Method | Links | |
|---|---|---|
| UniWhisperEvaluation Protocol=MLP2026.02 | 90.5 | |
| CLAP-UEvaluation Protocol=MLP2026.02 | 87.6 | |
| Concat.Encoder Type=Fusion2026.06 | 83.8 | |
| WQ-FusionEncoder Type=Fusion2026.06 | 82.8 | |
| Gated Trans.Encoder Type=Fusion2026.06 | 82.1 | |
| BEATsEvaluation Protocol=MLP2026.02 | 79.1 | |
| Adapt. and Trans.Encoder Type=Fusion2026.06 | 78.7 | |
| Qwen2-Audio-7BEncoder Type=Single2026.06 | 76.6 | |
| Ensemble (Submission)Ensemble Strategy=Embedding Concatenation2026.01 | 76.4 | |
| Whisper-LargeEncoder Type=Single2026.06 | 71.9 | |
| BEATs (300M) BalancedParameter Count=300M, Pre-training Mixture=Balanced (40:30:30)2026.01 | 68.9 | |
| AudioMAEEncoder Type=Single2026.06 | 68.2 | |
| Dasheng 1.2BParameter Count=1.2B2026.01 | 62.7 | |
| BEATs (300M) SpeechParameter Count=300M, Pre-training Mixture=Speech-heavy (70:15:15)2026.01 | 61.2 | |
| Whisper-U-3Evaluation Protocol=MLP2026.02 | 55.8 | |
| Challenge BaselineDescription=Best among Dasheng-base, data2vec, and Whisper2026.01 | 55.7 | |
| BEATs (90M) iter3Parameter Count=90M, Iteration=32026.01 | 54.5 | |
| Dasheng-BaseEncoder Type=Single2026.06 | 41.5 | |
| WavLMEvaluation Protocol=MLP2026.02 | 36.4 | |
| WhisperEvaluation Protocol=MLP2026.02 | 36 | |
| HuBERTEvaluation Protocol=MLP2026.02 | 34.9 | |
| wav2vec 2.0Evaluation Protocol=MLP2026.02 | 23.1 | |
| Whisper-UEvaluation Protocol=MLP2026.02 | 21.8 | |
| CLAPEvaluation Protocol=MLP2026.02 | 6.4 | |
| CLAP-UEvaluation Protocol=kNN2026.02 | 0 |