Anthology of Computers and the Humanities · Volume 4

Les formats d’encodage de la notation musicale à l’épreuve d’un objectif de transcription manuscrite automatique

Hugo Scheithauer1 , Gonzalo Romero-García2 , Laurent Romary1 et Thibault Clérice1

  • 1 Inria, ALMAnaCH, Paris, France
  • 2 EPITA Research Laboratory, Image Processing and Pattern Recognition, Paris, France
Download PDF Download Citation

Permanent Link: https://doi.org/10.63744/K2fvu4Nzsv6t

Published: 21 May 2025

Keywords: optical music recognition, handwritten music recognition, transcription, dataset, standards, common western music notation

Mots clés : reconnaissance optique de musique, transcription automatique de partitions manuscrites, transcription, jeu de données, standards, notation musicale occidentale

Abstract

The development of Optical Music Recognition (OMR) faces fragmentation in representation formats, divided between editorial requirements (MEI), demands for graphical rendering and semantics (MusicXML), and computational analysis (Humdrum **kern). This tension is particularly evident in handwritten sources, where musical semantics are coupled with a complex documentary materiality. Building on recent advances in ATR (Automatic Text Recognition), this article evaluates the ability of current standards to serve as ground truth. Using Claude Debussy’s piano manuscripts as a case study, we advocate a graphematic approach to transcription. We justify repurposing MusicXML toward a “structural” model capable of preserving, to some extent, the physical anchoring and non-standard phenomena of the source. This proposal aims to contribute to the stabilization of annotation practices for the creation of shareable datasets, an essential condition for the development of robust historical OMR.