Randomized Scores and Diverse Timbres: Augmenting Automatic Music Transcription with Online-Generated Data
Published in arXiv preprint, 2026
Recommended citation: Haiwen Xia, Chao Zhang, Qiuqiang Kong. (2026). "Randomized Scores and Diverse Timbres: Augmenting Automatic Music Transcription with Online-Generated Data." arXiv preprint arXiv:2610.11197. https://arxiv.org/abs/2610.11197
An online sampler-renderer pipeline separates score structure from instrument timbre to study which synthetic examples help automatic music transcription generalize. Moderate note-event randomization remains useful, while diverse timbres provide consistent gains across domains.
| Paper | Code |