Proceedings of the 4th International Workshop on Reading Music Systems

Abstract: The International Workshop on Reading Music Systems (WoRMS) is a workshop that tries to connect researchers who develop systems for reading music, such as in the field of Optical Music Recognition, with other researchers and practitioners that could benefit from such systems, like librarians or musicologists. The relevant topics of interest for the workshop include, but are not limited to: Music reading systems; Optical music recognition; Datasets and performance evaluation; Image processing on music scores; Writer identification; Authoring, editing, storing and presentation systems for music scores; Multi-modal systems; Novel input-methods for music to produce written music; Web-based Music Information Retrieval services; Applications and projects; Use-cases related to written music. These are the proceedings of the 4th International Workshop on Reading Music Systems, held online on Nov. 18th 2022.

• The paper introduces a novel OMR dataset featuring 19th-century musical examples that challenges conventional training sources.
• The paper demonstrates the application of deep neural networks, like CompIdNet, to accurately identify composers from sheet music.
• The paper presents robust end-to-end OMR systems using score unfolding and graph-based predictions that advance complete music transcription.

Overview of the 4th International Workshop on Reading Music Systems (WoRMS) Proceedings

The 4th International Workshop on Reading Music Systems (WoRMS) compiled a diverse collection of nine research papers that explore innovative methodologies and technologies in the domain of Optical Music Recognition (OMR) and related systems. Held online in 2022, this workshop served as a confluence for researchers worldwide who share a vested interest in developing systems to interpret and analyze musical scores accurately.

The array of topics covered within this workshop's proceedings reflects the breadth of current research in music recognition systems. These topics encompass dataset generation, novel approaches for music notation assembly, measure detection, and the transcription of electronic drum kits, among others.

Key Contributions

1. Datasets for OMR: The introduction of a new dataset aimed at challenging the conventional sources in OMR was presented by Moss, López, Köster, and Rizo. This dataset incorporates diverse examples of 19th-century music theory, pushing the boundaries of how machine learning models can learn to recognize historical and often complex notations.
2. Composer Identification through DNNs: Walwadkar et al. propose CompIdNet, which employs Deep Neural Networks to identify composers from sheet music. This contribution underscores the effectiveness of deep learning algorithms in deciphering stylistic fingerprints unique to different composers.
3. Challenges in Synthesizing Training Data: Mayer and Pecina's work outlines the obstacles faced in generating training data for OMR systems. Their research emphasizes the necessity of overcoming these challenges to improve the accuracy and generalization of OMR models.
4. End-to-End OMR Systems: The works of Ríos, Iñesta, and Calvo-Zaragoza, as well as Garrido-Munoz et al., both push forward the concept of end-to-end systems in OMR. The first focuses on monophonic documents via score unfolding, while the latter investigates graph-based predictions, offering distinct methods in text-free music recognition pipelines.
5. Notation Assembly and Measure Detection: Efficient approaches for assembling music notations, as discussed by Penarrubia et al., and the development of computer-assisted measure detection systems as proposed by Egozy and Clester, mark significant advances in making OMR systems more robust and pragmatic for real-world applications.
6. Drum Transcription and Handwritten Music: Jacquemard et al.'s automated transcription of electronic drum kits, alongside Torras et al.'s integration of LLMs in enhancing handwritten music recognition, highlight the ongoing efforts to broaden the scope of musical elements and styles that can be effectively captured by modern OMR systems.

Implications and Future Directions

This compilation of research contributions reaffirms that while substantial progress has been made in the domain of reading music systems, ongoing challenges remain, particularly in dealing with varied notational conventions and historical documents. The performance of music recognition systems can vastly benefit from enhanced datasets that encapsulate a wider range of musical styles and representations.

Practically, these contributions pave the way for more sophisticated applications in digital musicology, music information retrieval, and interactive music education tools. As OMR technologies continue to integrate more complex datasets and utilize advanced neural architectures, they hold promise for significantly transforming how music is interpreted and consumed in digital formats.

Future research is likely to expand upon these findings, potentially exploring more robust combinations of symbolic music representations and harnessing emerging AI technologies such as transformers in the domain of music recognition. The progression towards fully automated systems that can seamlessly handle the vast complexities of music notation is anticipated to remain a critical area of development in artificial intelligence and computational musicology.