Subjective Emotional Alignment on Story2MIDI (Listening Study)
0.53Valence AccuracyStory2MIDI
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Story2MIDIEncoder=RoBERTa, Decoder=Transformer (3 layers, 4 heads), Training=Contrastive Training2025.12 | 0.53 | 0.7 | 0.4 |
Loading the SOTA2 catalog…