Score-informed Piano Transcription Dataset

Score-informed Piano Transcription Dataset

Emmanouil Benetos 2012

All pieces were recorded using an untuned Yamaha U3 Disklavier at the Media and Arts Technology Lab, School of Electronic Engineering and Computer Science, Queen Mary University of London, UK. The dataset contains recordings with some performance mistakes, as well as midi annotation of all the mistakes made by the performer.

Each piece is linked to the following files:
1) Disklavier_*.wav: recording of the piece, containing mistakes
2) Disklavier_*_correct.mid: midi file with correct notes, aligned to the .wav performance
3) Disklavier_*_missed.mid: midi file containing the missed notes, not present in the recording (aligned)
4) Disklavier_*_fa.mid: midi file containing the extra played notes (false alarms) in the recording (aligned)
5) Score_*.mid: midi file with the non-aligned original score of the piece

Piece Information (from ABRSM 2011/12 syllabus, grades 1 and 2):
01 Josef Haydn Andante: from Symphony No. 94 in G, Hob. I/94, second movement
02 James Hook Gavotta: No. 3 from 24 Progressive Lessons, Op. 81
03 Pauline Hall Tarantella
04 Felix Swinstead A Tender Flower
05 Johann Krieger Bouree: from Sechs musicalishe Partien
06 Johannes Brahms The Sandman: No. 4 from Volks-Kinderlieder, WoO 31
07 Trad. American Down by the Riverside

The dataset also includes 4 sets of chromatic scales recorded from the Disklavier ('Disklavier_ChromaticX.wav').

==================================================================================================

The download section contains two ZIP files. The first file contains the original database as published with the following paper in 2012:
Emmanouil Benetos, Anssi Klapuri and Simon Dixon. "Score-informed transcription for automatic piano tutoring", Proceedings of the European Signal Processing Conference (EUSIPCO), 2012.

A second ZIP file contains a modified/corrected set of 'ground truth' annotation file that were made available in 2016 as part of the following publication:
Sebastian Ewert, Siying Wang, Meinard Müller and Mark Sandler. "Score-Informed Identification of Missing and Extra Notes in Piano Recordings", Proceedings of the International Society for Music Information Retrieval Conference (ISMIR), 2016.