kkwsts/MAJEPPA-Dataset

MAJEPPA

4,449 solo piano performance MIDI files (279 hours, ~5.0M notes) transcribed from amateur and professional piano recordings, each labelled with the performer's expertise level and recording context.

• 4,207 (95%) paired with a reference score in MIDI — 886 scores covering 880 distinct works and movements by 52 composers
• 3,862 (87%) carry a DTW score↔performance alignment — 20 million timestamp correspondences
• Every performance links to its source YouTube recording

The corpus covers the full skill spectrum.

No audio is distributed. Symbolic MIDI, metadata, and alignments only.

Contents

File	Contents
performance.zip	P_0001.mid … P_4449.mid — 4,449 performance transcriptions
score.zip	S_0001.mid … S_0886.mid — 886 reference scores
metadata.parquet	4,449 rows × 12 columns, one per performance
metadata.csv	the same table as CSV, for convenience
alignment.parquet	20,048,645 rows — DTW warping paths

metadata.csv

Column	Description
performance_id	P_NNNN, file in performance.zip
recording_id	R_NNNN, groups clips cut from the same source video
score_id	S_NNNN, file in score.zip. Empty when unmatched
composer	Canonical composer name. Empty for 137 rows
piece_title	Work title, including movement where applicable
recording_type	practice, demo_class, performance, concert_performance, sight_read, slow_demo, fast_demo
expertise_level	piano_teacher, adult_beginner, adult_intermediate, virtuoso, child_beginner, child_professional
clip_start_s	Offset of this clip within its source video
clip_end_s	End offset. Empty means "to the end of the video", not missing
alignment_cost	DTW alignment cost
score_coverage	Fraction of the score matched
youtube_url	Source video (the link might be invalid or removed)

performance_id and score_id are independent ID spaces. One score serves many performances; use score_id for the pairing.

alignment.parquet

performance_id, score_s, perf_s (float32), in chronological order. Each row is one point on the DTW warping path: a score timestamp and the performance timestamp it maps to. Rows are in chronological order. Join to metadata.csv on performance_id.

Statistics

	Prac.	Sight.	Perf.	Demo*	Slow	Conc.	Total	Hours
Piano teacher	35	15	56	771	329	201	1,407	72.1
Adult beginner	637	404	94	17	3	1	1,156	112.6
Adult intermediate	464	140	175	22	2	3	806	45.7
Child beginner	188	8	197	5	2	0	400	8.9
Child professional	90	0	183	3	0	0	276	13.3
Virtuoso	0	0	0	0	0	404	404	26.7
Total	1,414	567	705	817	336	609	4,449	279.3

Ethical considerations

Performances originate from publicly published YouTube videos for research purposes under fair dealing/use. We release only derived MIDI and coarse descriptive tags (expertise and recording context), not raw audio or personally identifying metadata beyond what is already public, and we will respond promptly to any request from an uploader, performer, or rights holder to remove specific content from the dataset. Additionally, we intend the resource for research on performance understanding and music education rather than for replacing teachers or performers. 

Citation

If you use this dataset, please cite:

@inproceedings{zhou2026majeppa,
title  = {MAJEPPA: Morphing and Assessing in a Unified Piano Performance Space},
author = {Zhou, Jinwen and Zhang, Huan and Zhai, Weixi and Liang, Jinhua and Hogg, Aidan O. T. and Dixon, Simon},
booktitle = {Proc. International Society for Music Information Retrieval Conference (ISMIR)},
year   = {2026}
}

Contact

jinwen.zhou@qmul.ac.uk