| title | SheetSage2: Coherent Lead-Sheet Transcription with Synthetic Supervision |
|---|---|
| authors | Junyan Jiang, Ruibin Yuan, Jiahao Pan, Wei Xue, Yike Guo, Gus Xia, Yann LeCun |
| date | 2026-10-04 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2610.05336 |
| affiliation | Multimodal Art Projection |
| upvotes | 13 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf |
| abstract | Transcribing music into a human-readable score requires a coherent understanding of rhythm, harmony, melody, and form. Two obstacles limit this goal: annotated recordings are scarce, and accurate local predictions can still produce inconsistent musical sequences. We present SheetSage2, a unified music transcription framework that combines synthetic data, task-specific structured decoding, and autoregressive distillation. Automatically annotated MIDI, rendered into audio, provides scalable supervision across music understanding tasks. Task-specific structured decoders integrate complementary musical cues and their temporal dependencies to produce musically coherent scores. Autoregressive distillation further retains transcription accuracy without task-specific dynamic programming at inference. Across eight benchmark collections, a single SheetSage2-AR model exceeds the listed prior systems on 12 of 15 benchmark--metric pairs in our evaluation, substantially improving over SheetSage1 and surpassing task-specific models on several benchmarks. Model weights and inference code are publicly available. |
cite
@misc{jiang2026sheetsage,
title={SheetSage2: Coherent Lead-Sheet Transcription with Synthetic Supervision},
author={Junyan Jiang and Ruibin Yuan and Jiahao Pan and Wei Xue and Yike Guo and Gus Xia and Yann LeCun},
year={2026},
eprint={2610.05336},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2610.05336}
}