Timbre-agnostic transcription on BACH10 (tutti)
86.5F_F ScorelCQTMN
Evaluation results
| Method | Links | ||
|---|---|---|---|
| lCQTMNTraining Dataset=MusicNet, Feature Extraction=Learnable CQT2025.09 | 86.5 | 79.2 | |
| BPflTraining Dataset=Synthetic, Loss Function=Focal Loss2025.09 | 86.1 | 75.2 | |
| BNTraining Dataset=Synthetic, Normalization Strategy=BatchNorm2025.09 | 85.9 | 76.4 | |
| BPflMNTraining Dataset=MusicNet, Loss Function=Focal Loss2025.09 | 85.7 | 80.5 | |
| MNTraining Dataset=MusicNet2025.09 | 84.8 | 79.4 | |
| BasicPitchTraining Dataset=Synthetic2025.09 | 84.8 | 53.7 | |
| Lightweight Two-Branch ArchitectureTraining Dataset=Synthetic, Normalization Strategy=EnergyNorm, Convolution Type=Dilated2025.09 | 84.6 | 75.2 | |
| OFflMNTraining Dataset=MusicNet, Loss Function=Focal Loss2025.09 | 84.6 | 76.7 | |
| BPlossTraining Dataset=Synthetic, Loss Function=BasicPitch's loss2025.09 | 84.3 | 63.6 | |
| Conv39Training Dataset=Synthetic, Convolution Type=39-tap2025.09 | 84.1 | 74.6 | |
| Onsets&FramesTraining Dataset=Synthetic, Loss Function=BCE loss2025.09 | 82.7 | 70.7 | |
| OFflTraining Dataset=Synthetic, Loss Function=Focal Loss2025.09 | 82.1 | 72.3 | |
| noLogTraining Dataset=Synthetic, Normalization Strategy=EnergyNorm without log2025.09 | 80.7 | 68.1 | |
| lCQTTraining Dataset=Synthetic, Feature Extraction=Learnable CQT2025.09 | 79.1 | 63.5 |