PawCT transcribes choral audio into separate SATB note tracks
Original title: Toward Part-Aware Choral Transcription with singing voice assignment
Authors report a 0.225 part-aware note F1, above an adapted baseline and a two-stage assignment pipeline.
Researchers introduce Part-aware Choral Transcription, an end-to-end neural system that identifies active soprano, alto, tenor, and bass parts and writes a separate note track for each. It combines part-specific onset, offset, and frame prediction with part-presence estimation and pitch-range priors. On YouChorale, the authors report PawCT-RP-OC ahead of prior choral transcription approaches, with further tests on CSD and Cantoria.
Unconfirmed: Scores are the authors’ reported results; no independent replication is described.
- PawCT-RP-OC macro part-aware note F1 is a reported 0.225 at 50 ms.
- An adapted choral baseline scores 0.165 and a two-stage pipeline 0.175.
- The part-agnostic PagCT variant reaches a reported 0.382 onset F1 versus 0.237.
- Code and demos are linked from the project page.
From the original article
1 min read at huggingface.co
Single-instrument automatic music transcription (AMT) has advanced substantially, yet choral applications require soprano, alto, tenor, and bass (SATB) to be transcribed as separate parts. Recent note-level choral AMT instead produces a single merged note track, limiting rehearsal, education, and score reconstruction.
By Hanyu Meng, Zhanhong He, Zixun Guo · 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.