Advancing Music Processing for Concert Band and Wind Music

• Applicant: Professor Dr. Meinard Müller
• Subject Area: Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 555525568

This project aims to advance music processing in the context of concert band and wind music, a domain deeply integrated into Germany's cultural landscape yet significantly overlooked in the research field of Music Information Retrieval (MIR). Focusing on brass and woodwind instruments across a broad musical range, from classical to contemporary compositions, we aim to explore extensive possibilities for data-driven music and audio processing research. In particular, the project addresses two main objectives: achieving technical advancements in MIR and applying these advancements within a culturally significant amateur music context. On the technical research side, our general objective is to expand the scope of current MIR techniques, which have largely focused on mainstream popular music or data-rich scenarios like piano music. We aim to explore the unique realm of concert band and wind music, systematically evaluating, refining, and customizing data-driven MIR methods. Our specific areas of focus include automatic music transcription, instrument recognition, music source separation, and generative music sound synthesis. As part of our overarching conceptual strategy, we will study suitably tailored loss terms and conditioning techniques to account for prior musical knowledge. Furthermore, we will investigate multi-task strategies for multi-instrument music transcription, incorporating analysis-by-synthesis approaches based on differentiable digital signal processing and source separation techniques. In this context, one main goal is to investigate, understand, and mitigate the impact of confounding factors, such as co-occurring instruments, instrument-specific musical motifs, or recording conditions, that cause models to respond to patterns not directly related to the desired musical or acoustical attributes. On the music application side, our objective is to investigate how advanced MIR techniques can transform music education and rehearsal practices within amateur concert band and wind music communities. This effort encompasses more than just data organization and collection; it includes creating tools to help amateur musicians become familiar with the repertoire and to spark enthusiasm for this music among new generations. Beyond these primary objectives, the project adopts open-source principles and fosters collaborations between engineers, computer scientists, musicians, conductors, and amateur concert bands. These collaborations will enable us to create music test sets and conduct real-world evaluations. Consequently, our holistic approach aims to drive substantial progress in MIR research by delving into a seldom-explored musical landscape, while supporting cultural heritage and enhancing music education.

• DFG Programme: Research Grants
• Co-Investigator: Dr.-Ing. Stefan Balke