ANR-FRQSC-2 - AAP franco-québécois en Sciences humaines et sociales

Music Notation Information Retrieval – MuNIR

MuNIR, towards search engines applied to collections of digital scores

The MuNIR project aims to design and validate new tools to manage and manipulate digital musical scores. The potential of this digitization remains dependent on the lifting of several locks. The scientific obstacles identified and targeted by the project: mass digitization, the ability to manage large collections of partitions (interrogation) in a scalable manner (indexing, optimization) and the development of generic methods of analysis.

Digitisation of historical scores and the promotion of their content: towards intelligent digital music libraries

Music scores archives constitute an important part of cultural heritage. They are present in all the major heritage libraries such as the National Library of France. Digitisation of multimedia content is actively undertaken by these institutions. When applied to score collections, they result in images whose content is difficult to handle due to the particular nature of music notation. In addition, search and interaction tools, again very specific to the nature of the content represented, are hardly possible on the basis of an image, due to the lack of techniques and methodologies adapted and robust enough. The MuNIR project sought to remove these obstacles by studying mass digitisation, manage tools dedicated to large collections of partitions (query, indexing, optimisation) and the development of generic analytic methods.

MuNIR relies on new formats for digitised scores which have emerged recently and allow very fine representation of music notation. This content is very diverse: it features explicit information (notes, instruments, chords), metadata (authors, dates), layout instructions, as well as implicit information relating to the musical language. MuNIR has worked to promote the MEI encoding format, developing methods for converting from image format to MEI («optical music recognition », OMR) format and to exploit this format. The basic building blocks of a search engine have been developed: query language, indexing structures, and knowledge extraction. In order to provide support and recommendations to the scholar community most concerned by the project, data sets and software components are made available. All of these results constitute a significant advance towards the ability to provide researchers, musicians and archivists with efficient and scalable research and analysis tools.

The project studied the digitisation process that encodes musical notation from the most common multimedia sources: images and audio. In addition to scientific publications on these two subjects, the project partners have created an annual workshop, WORMS, the first edition of which was held at Cnam in 2018 (https://sites.google.com/view/worms2018/home) and the third of which will take place in Montreal in 2020. We have also produced and made available to the community annotated data sets which can serve as a training sets and currently constitute the most complete source in the field of noted music (by opposition with audio documents). Finally, Open Source tools for managing large collections of digitised partitions are also produced and supported by the partners.

The project established a solid foundation for modeling digital partitions and extracting knowledge. On these bases, the most important future works raise the challenge of large-scale digitization to allow the availability of the very vast heritage currently existing in the form of images. The partners will focus on their respective areas of expertise: automatic transcription and optical recognition.

Both partners are contributors to the Music Encoding Initiative, whose annual conference was organised in France (in Tours) in 2017, under the co-direction of P. Rigaux (Cnam) and P. Vendrix (Tours). We place all of our work in the context of promoting this initiative and its encoding format. As such we have published in all specialised conferences (TENOR, ISMIR, DLFM, MCM). The query language defined in the project was the subject of a publication in a main generalist journal in computer science, Information Systems. Finally, the two partners endeavoured to contribute to the software tools and their distribution via their respective platforms, GitHub and Music21.

The MuNIR project aims at contributing to SSH research in the digital age by serving the search & analysis needs of humanities scholars who work on music notation, musical heritage, and the analysis of musical language as it changes over time. To this end, we plan to conduct a coordinated research program devoted to the search and analysis of music that has been encoded into representative symbolic music format (scores), and to the development of collections and tools to demonstrate the practical impact of our work and to disseminate our results. The project aims to answer two closely related questions:
1) Given a very large collection of music scores, how can we organize it to make its access easier, faster? How can we find and communicate the structure and principles of the music language implicitly encoded in its notation? What methodologies and tools can help to make sense of large and complex music scores ?
2) How can we make our research achievements usable by the SSH community in order to leverage their collections of digitized music scores, as well as their studies on music heritage?

To address the first question, we will develop a research activity focused on information retrieval and analysis challenges: This covers structures and algorithms for searching (including by content), extraction of structural or semantic information, and production of high-level descriptors (metadata) that will help to group similar music scores, to classify them according to genre, style or composer, or to discover recurring patterns.
This research endeavour will be completed by a development and dissemination effort to address the second question. Our goal is to design and implement a framework providing a reference architecture for designing search & analysis platforms on large collections of music scores, along with open source software packages to quickly implement such systems. This will strongly encourage the creation of applications to provide musicologists all over the world with easy access to large collections and interfaces that support sophisticated analytic methods.

Project coordination

Philippe Rigaux (Conservatoire National des Arts et Métiers)

The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.

Partner

CNAM Conservatoire National des Arts et Métiers
McGill McGill University

Help of the ANR 211,255 euros
Beginning and duration of the scientific project: December 2016 - 36 Months

Useful links

Explorez notre base de projets financés

 

 

ANR makes available its datasets on funded projects, click here to find more.

Sign up for the latest news:
Subscribe to our newsletter