Timbre-separated transcription on BACH10 4mix
68.5F-score (Frame-based)Note-Level Contrastive Clustering
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Coupling Strategy=Shared residual block, Loss=InfoNCE [8]2025.09 | 68.5 | 72.8 | 91 | |
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Loss=InfoNCE [8]2025.09 | 66.2 | 72.7 | 89.9 | |
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Embedding Dimension=16, Loss=InfoNCE [8]2025.09 | 64.5 | 70.4 | 87.1 | |
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Coupling Strategy=Rescale, Loss=InfoNCE [8]2025.09 | 63.6 | 70.9 | 87.7 | |
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Loss=L_affinity [6]2025.09 | 59 | 67.9 | 84 | |
| TanakaBaseline=Tanaka et al. (2020), Loss=InfoNCE [8]2025.09 | 55.4 | 59.2 | 73.2 | |
| Note-Level Contrastive ClusteringTraining Data=Synthetic, Loss=InfoNCE [8]2025.09 | 46 | 56.1 | 69.4 |