SheetSage2: Coherent Lead-Sheet Transcription with Synthetic Supervision
SheetSage2 transcribes coherent lead sheets using synthetic MIDI supervision and beats prior systems on most reported metrics.
SheetSage2 is a music lead-sheet transcription system that combines synthetic supervision, structured decoding, and autoregressive distillation. Automatically annotated MIDI is rendered to audio to scale training across music-understanding tasks, while task-specific decoders enforce temporal musical consistency. Distillation keeps accuracy without dynamic programming at inference. On eight collections, one SheetSage2-AR model beats listed prior systems on 12 of 15 benchmark-metric pairs and improves on SheetSage1; weights and inference code are released.
- Uses rendered, auto-annotated MIDI as scalable supervision.
- Structured decoders enforce rhythm, harmony, melody, and form.
- Autoregressive distillation removes inference-time dynamic programming.
- SheetSage2-AR wins 12 of 15 listed benchmark-metric pairs.
- Weights and inference code are public.
- AI models
- SheetSage2SheetSage2-ARSheetSage1
From the original article
1 min read at huggingface.co
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.
By Junyan Jiang, Ruibin Yuan, Jiahao Pan · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.