Timbre-separated transcription on BACH10 3mix
77.8F-score FSNote-Level Contrastive Clustering
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Loss=InfoNCE [8]2025.09 | 77.8 | 80.1 | 96.7 | |
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Embedding Dimension=16, Loss=InfoNCE [8]2025.09 | 76.8 | 79.5 | 96 | |
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Coupling Strategy=Rescale, Loss=InfoNCE [8]2025.09 | 76.8 | 79.6 | 96.2 | |
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Coupling Strategy=Shared residual block, Loss=InfoNCE [8]2025.09 | 76.1 | 78.9 | 96.6 | |
| Note-Level Contrastive ClusteringTraining Data=MusicNet, Loss=L_affinity [6]2025.09 | 70.8 | 75.7 | 91.4 | |
| TanakaBaseline=Tanaka et al. (2020), Loss=InfoNCE [8]2025.09 | 66.1 | 69 | 83.4 | |
| Note-Level Contrastive ClusteringTraining Data=Synthetic, Loss=InfoNCE [8]2025.09 | 58.7 | 68 | 82.1 |