Timbre-separated transcription on BACH10 2mix
83.4F-score (Frame-based Separation)Note-Level Contrastive Clustering
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Loss=InfoNCE [8]2025.09 | 83.4 | 84.6 | 99 | |
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Coupling Strategy=Rescale, Loss=InfoNCE [8]2025.09 | 83.4 | 84.6 | 99.1 | |
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Embedding Dimension=16, Loss=InfoNCE [8]2025.09 | 83.2 | 84.4 | 98.8 | |
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Coupling Strategy=Shared residual block, Loss=InfoNCE [8]2025.09 | 82.1 | 83 | 98.4 | |
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Loss=L_affinity [6]2025.09 | 80.4 | 82.2 | 96.3 | |
| TanakaBaseline=Tanaka et al. (2020), Loss=InfoNCE [8]2025.09 | 77.9 | 79.6 | 93.2 | |
| Note-Level Contrastive ClusteringTraining Data=Synthetic, Loss=InfoNCE [8]2025.09 | 72.4 | 78.1 | 91.4 |