Development of an AI-based, level-oriented repertoire platform for music teachers

Comprehensive digital sheet music libraries or collections of teaching materials designed to support teachers in lesson planning have so far found little traction in instrumental and vocal tuition. In their project, Kerstin Denecke and Andrea Ferretti are investigating which criteria and features a repertoire platform would need to fulfil in order to offer added value to music teachers. In doing so, they are making a valuable contribution to promoting the use of digital resources in the arts and supporting the digital literacy of music teachers.

Repertoire selection as a central task for teachers in music education

Selecting a suitable repertoire tailored to learners’ abilities and learning objectives is a central yet time-consuming task for teachers at music schools and conservatoires (Jordhus-Lier et al., 2023; Nielsen et al., 2023). Music educators must continually identify compositions that match the technical abilities, musical goals, and individual characteristics of their learners. Although many large digital sheet music libraries have emerged in recent years, existing platforms rarely offer structured metadata or search functions tailored to the needs of music educators.  Platforms should therefore provide the ability to filter compositions according to precise didactic criteria—such as technical requirements (Ramoneda et al., 2022), learnable techniques, and instrumentation (Taenzer et al., 2019; De Pasquale et al., 2020)—or to receive targeted suggestions for comparable or more advanced pieces. The use of artificial intelligence (AI) methods could open up new possibilities (Cheng, 2019) for finding suitable compositions while also supporting evidence-based recommendations.

Which criteria are most important to teachers?

In a networking project between research and teaching funded by the BFH thematic field «Human Digital Transformation», we investigated the criteria teachers use to select compositions for lessons and how these can be incorporated into an AI-driven model. To this end, we interviewed experts and surveyed 37 instrumental and vocal teachers with varying levels of teaching experience using a questionnaire. This enabled us to derive empirically grounded requirements for a model that supports teachers in selecting compositions tailored to the individual needs of their learners.

The results show that the music teachers surveyed consider educational and practical aspects to be particularly relevant selection criteria. Information regarding the context of a composition was rated as significantly less important. The highest levels of agreement were recorded for instrumentation (very important: 59.5%), the technical level required to perform the piece (59.5%), and the skills that can be developed through the piece (40.5%). Similarly, whether a piece is written for a solo instrument (with or without accompaniment) or for an ensemble was regarded as important or very important by the majority of respondents. Overall, the results indicate that the highest-rated criteria are those directly related to adapting the repertoire to a learner’s technical abilities and educational needs.

First recommendation system: Language-based queries for repertoire

Based on these findings, we developed a model that enables language-based user queries for a music repertoire platform. The tested model is based on a large language model, which analyses user queries. For example, a user might ask: “Which contemporary piece is suitable for an intermediate-level flutist to strengthen flexibility in embouchure?” In a first step, the model analyses the features contained in the query (instrument = flute, level = intermediate, skills = embouchure, style = contemporary). The system then searches the stored repertoire for suitable pieces and presents the best matches.

As a web-based application, such a conversational assistant could help teachers quickly find appropriate pieces. The developed model could be adapted for use with existing libraries and platforms. We anticipate that, as digitalisation progresses, electronic libraries and platforms containing sheet music and other teaching materials will continue to expand. The use of tools for the targeted selection of learning materials tailored to individual learners could contribute to improving teaching quality as well as broadening the range of materials used in instrumental and vocal tuition.

References

Cheng, L. (2019). Musical competency development in a laptop ensemble. Research Studies in Music Education, 41(1), 117–131. https://doi.org/10.1177/1321103X18773804

De Pasquale, G., Spahiu, B., Ducange, P., & MAurino, A. (2020). Towards automatic classification of sheet music. CEUR Workshop Proceedings, 2646, 266–277.

Jordhus-Lier, A., Karlsen, S., & Nielsen, S. G. (2023). Meaningful approaches to content selection and ways of working: Norwegian instrumental music teachers’ experiences. Frontiers in Psychology, 14, 1105572. https://doi.org/10.3389/fpsyg.2023.1105572

Nielsen, S. G., Jordhus-Lier, A., & Karlsen, S. (2023). Selecting repertoire for music teaching: Findings from Norwegian schools of music and arts. Research Studies in Music Education, 45(1), 94–111. https://doi.org/10.1177/1321103X221099436

Ramoneda, P., Tamer, N. C., Eremenko, V., Serra, X., & Miron, M. (2022). Score difficulty analysis for piano performance education based on fingering (Version 1). arXiv. https://doi.org/10.48550/ARXIV.2203.13010

Taenzer, M., Abeßer, J., Mimilakis, S., Weiß, C., Müller, M., & Lukashevich, H. (2019). Investigating CNN-Based Instrument Family Recognition for Western Classical Music Recordings [Dataset]. Zenodo. https://doi.org/10.5281/ZENODO.3258829