Introducing the TISMIR Education Track: What, Why, How?

1.1 Education

Partly related to this ubiquity of music computing and the attendant access to music that it brings, MIR has not only emerged as a distinct research field but has also become increasingly attractive in education worldwide. Prominent universities now offer specialized courses in MIR and music processing across various disciplines such as computer science, engineering, and digital humanities.²

Music, with its ability to engage and fascinate people, serves as a powerful tool for interactive learning, providing tangible examples, hands-on exploration, and opportunities for experimentation (Müller et al., 2021). The interdisciplinary nature of MIR exposes students to a wide range of subjects, fostering a holistic understanding of both technical and cultural aspects. Furthermore, the practical applications of MIR, including music recommendation systems and audio transcription, resonate with learners, enhancing their comprehension and skill development through real-world applications.

As MIR gains prominence in research, educational literature on the subject is gradually becoming more available, although the pace of this growth remains relatively slow. Examples include textbooks on MIR topics in recent years such as those by Müller (2007); Müller (2015, 2021), Lerch (2012, 2022), Weihs et al. (2016), and Knees and Schedl (2016). Moreover, many articles published in the ISMIR proceedings and other venues often meet a high educational standard, offering material suitable for teaching and learning specific MIR topics. Insightful educational discussions on specific MIR topics are frequently present in introductory chapters of PhD theses published in the MIR field. Furthermore, ISMIR regularly hosts several tutorials at its annual conferences, which are well-attended and warmly received by the ISMIR community.³ Educational materials are also available in informal blogs or interactive programming notebooks provided on platforms like GitHub.

However, most of these educational resources on MIR research are scattered across platforms and venues, making them difficult to locate. Additionally, some sources do not guarantee a high standard of quality, as they do not employ any review process and so lack feedback mechanisms by MIR experts for quality control. As such, each prospective user must vet them from scratch. Finally, even well-intentioned users face challenges in properly referencing these materials due to factors such as the lack of academic publication, ambiguous licensing, and unclear authorship and origins.

Addressing many of these issues, a peer-reviewed and academically published educational article holds a distinct place in the educational ecosystem. It is particularly suitable for providing supportive information (e.g., ‘what’ and ‘how to’ for non-routine tasks) and procedural information (e.g., ‘how to’ for routine tasks), while accompanying notebooks or websites can complement these materials for practicing tasks (van Merriënboer, 2019).

1.2 TISMIR as a Venue for Education

To meet the growing need for high-quality educational resources in MIR, we have established an education track within the TISMIR journal. This track provides a dedicated platform for disseminating educational content tailored to the needs of students, educators, and researchers in MIR. Our aim is to enhance the accessibility, credibility, and impact of educational materials in the field by offering structured, peer-reviewed articles. Educational articles in TISMIR should systematically introduce fundamental theories, common practices, and/or applications in well-defined areas of interest to readers across various MIR-related fields, whether those fields are mature or emerging.

Examples may include specific signal processing or machine learning techniques, discussions on musicological or cognitive theories, introductions and reviews of recent MIR tasks, conceptual insights, or discussions on specific applications, reflecting the diverse and interdisciplinary interests of the MIR community. Through collaborative efforts among researchers, educators, and practitioners, our education track seeks to advance the pedagogical landscape of MIR and cultivate a community of lifelong learners dedicated to excellence in computational music research.

1.3 Organization of Editorial

In the following sections we reflect on what may constitute a well-crafted educational article. To achieve this we highlight key characteristics of educational articles in general and clarify review criteria (Section 2). Additionally, we explore why the music domain may provide an intuitive and motivating setting for education across various levels and disciplines (Section 3). Lastly, we provide examples of topics, concepts, and applications that could serve as an orientation aid for potential educational TISMIR articles (Section 4). In the concluding Section 5, we present the benefits of dedicating time and effort to a TISMIR education track submission. We wish to emphasize that this editorial is not intended to be comprehensive and should not constrain creativity or restrict the perspectives of potential authors. Instead, our objective is to offer guidance and foster discussions on crafting impactful educational articles for MIR and beyond.

2.1 Scope

The objective of a TISMIR educational article is to introduce an MIR-related subject in a clear and accessible way, accommodating a broad and interdisciplinary readership. An educational paper should systematically introduce fundamental theories, common practices, and applications in a well-defined, mature or emerging area, and preferably be of interest to readers across various MIR-related fields. A practical approach to defining the scope of an educational paper is to establish learning objectives, which are declarative statements that specify what learners are expected to know and do (Orr et al. 2022).

It is crucial to distinguish educational papers from conventional research articles, which typically present innovative techniques and novel findings. Educational articles should refrain from introducing new research or techniques, instead offering fresh perspectives on existing concepts and methodologies. In other words, educational articles should prioritize enhancing understanding rather than highlighting research innovations. Articles will succeed in making a novel contribution, not in the introduction of new techniques (for example), but in presenting a previously unavailable educational perspective on the subject matter. As a result, such an article may serve as a basis for a lecture in, e.g., information retrieval, machine learning, or signal processing. This perspective may also involve presenting a more STEM-oriented introduction to a subject usually taught in the humanities or social sciences. Conversely, it could involve examining the applicability of a specific engineering technique in addressing particular problems or needs within the humanities.

2.2 Comprehensibility/Clarity

Across various domains, educators would likely agree that effective educational articles should be accurate, clear, engaging, and relevant, regardless of the subject matter (Bloom and Engelhart, 1956; Krathwohl, 2002; Starr et al., 2008). A wishlist of key characteristics may include the following aspects:

• Accuracy. An educational article should contain accurate and current scientific information, ensuring that readers can trust the information provided and rely on it for learning purposes.

• Clarity. An educational article should provide clear and comprehensible explanations of scientific concepts, theories, and methodologies in a manner that is easily understandable.

• Jargon-Free. An educational article should be free of unnecessary jargon, and explain technical terms for readers with limited background knowledge.

• Links to Further Reading. While it may not always be possible to explain every concept from scratch, an educational article should provide clear links to external resources explaining any assumed knowledge. These links should be carefully vetted, open access if possible, and adopt the same educational standards outlined here.

• Engaging Presentation. An educational article should engage readers through an interesting presentation, including examples, visuals like diagrams or graphs, and interactive elements where possible. Real-world examples and analogies should be used to simplify complex concepts and improve understanding. If suitable, incorporating music examples in digital symbolic representations or as audio recordings can also be beneficial.

• Practical and Relevant. An educational article should demonstrate practical relevance by discussing how scientific concepts apply to concrete and real-world MIR applications, ensuring readers perceive the value of the information presented. Furthermore, it may illustrate how techniques and abstract concepts can be applied and scaled to solve problems in other domains beyond music.

Effective pedagogical strategies in educational articles, such as presenting the material in small steps, providing worked examples, summarizing key points, providing exercises for reflection, and suggesting further reading, are vital for facilitating learning (Rosenshine, 2012). Moreover, proper references and citations are essential for enhancing the credibility of content and facilitating deeper exploration of the topic for readers. They support claims made within the article and offer additional resources for further understanding and research. These are only a few key characteristics and suggestions. Additional insights are offered by Heard (2022) regarding writing high-quality scientific educational articles.

2.3 Intended Audience and Usability

An educational article should clearly state its purpose, audience, and how it can be used. This helps tailor the content and language accordingly. Clarifying its potential application in lectures, seminars, or practical settings also aids readers in understanding how to engage with the information. This ensures the article effectively serves its purpose and maximizes its impact on the intended audience.

The audience of an educational article can vary based on the subject matter and its intention. Generally, the audience comprises individuals interested in learning about a specific topic, including students, researchers, educators, and professionals. Educational articles are typically written to be accessible to readers with diverse levels of expertise, making them suitable for a broad audience.

When preparing an educational article for TISMIR, it may be helpful to envision it as providing a robust foundation for lectures, seminars, or lab sessions, targeting researchers new to the field, students, and teachers. For example, it could serve as comprehensive notes for a 60–90 minute lecture. Additionally, it should delve into specific techniques to assist researchers or practitioners in applying them to their respective research problems. Furthermore, when exploring an MIR task, it should encourage readers from diverse disciplines to explore potential applications within the music domain.

2.4 Resource-Related Principles

The TISMIR education track appreciates the FAIR principles of Findability, Accessibility, Interoperability, and Reusability (Wilkinson et al., 2016) and similar resource-related principles such as Availability and Reproducibility, and expects adoption of the same in submissions to the track.

Some of these principles are discussed by McFee et al. (2019) in the context of music signal processing research, and here we explicitly acknowledge the importance of these issues not only for research but also for educational articles. Availability ensures that the article is accessible to its intended audience, whether through open access publishing or through widely available platforms. Reproducibility refers to the ability to replicate the experiments or methods described in the article, which is crucial for validating the findings and ensuring that others can build upon the research. These aspects contribute to the credibility and usefulness of the educational article, allowing readers to access and verify the information presented.

In view of applicability and reproducibility, an educational article might offer links to test datasets, present a concise reference implementation of fundamental techniques, or incorporate supplementary multimedia materials like audio, images, sheet music, and symbolic data, if relevant. In this context, it is crucial that both the article and any supplementary material are open source, ensuring transparency, reproducibility, and trust. This fosters innovation and advancement in education, promoting equity and collaboration among researchers. Overall, open source and open access significantly enhance the dissemination, accessibility, and impact of educational materials, benefiting academia and society alike.

Connecting many of the above considerations is that of license. It is essential (and required of a TISMIR education track submission) that the associated code is not only open in a general sense but specifies the exact nature of that openness with a clear and explicit license. Anyone seeking to make secondary use of the materials needs to have the confidence that it is permitted and to know if they have to acknowledge the source (and how to do so). The same goes for any data involved in those materials. For example, the opacity of license has held symbolic MIR back (Gotham, 2021).

2.5 Review Process

Every educational article submitted to TISMIR will be vetted by the standard peer-review process, similar to any other TISMIR submission. However, different review criteria are employed compared to other article types. These criteria consider the key characteristics discussed before and include:

• Quality of language and presentation style.

• Emphasis on deepening understanding rather than research novelty.

• Introduction of a novel educational viewpoint that was previously unavailable.

• Appeal to a wide audience or to a well-defined target audience.

• Ease of reproducibility.

• Inclusion of instructive graphics, figures, and other multimedia materials.

• Consistency in mathematical and conceptual formalism where these are necessary.

We primarily support submissions that provide resources and educational materials on MIR. While TISMIR is not an educational journal, we remain open to other types of pedagogical papers addressing topics such as curriculum design, modes of assessment, diversity/inclusion, learning analytics, and plagiarism within the context of MIR.

All educational articles undergo a pre-screening by the TISMIR editorial board before being sent for peer review. Submissions may be rejected if they are out-of-scope or poorly written. However, authors of rejected submissions may be encouraged to consider submitting to the regular track if their submission contains novel research results. Authors may also be encouraged to resubmit after a thorough rewrite and restructuring, initiating a new review process.

Finally, it is important to note that the length of an educational submission should adhere to the same rules as regular TISMIR submissions, with a maximum of 8000 words. In exceptional cases where a paper is likely to exceed this limit, authors are encouraged to contact the journal to discuss available options.

3.1 Engagement and Motivation

Many people have a personal connection to music and are inherently motivated to learn about it. Using music as a context for teaching technical concepts can enhance engagement and motivation, leading to deeper learning and understanding. Furthermore, music encourages creativity and innovation, making it an exciting domain for exploring new ideas and approaches. For example, within the realm of generative music, interactive installations, or intelligent music composition systems, one can delve into the study of machine learning and signal processing techniques. Finally, music can be interactive, allowing users to actively engage with the content through activities such as playing instruments, composing, remixing tracks, or participating in live performances. Interactive music technologies enable users to manipulate and control musical elements in real-time, enhancing engagement and creativity.

3.2 Interdisciplinary Nature

The interdisciplinary nature of music and music processing spans a broad spectrum of fields and disciplines that intersect with the study and application of music. Below, we offer a selection of interdisciplinary connections to fields closely related to MIR.

• Music Theory. This field involves the study of the principles and structures underlying musical composition, which can include harmony, melody, rhythm, and form (Eerola, 2024). Thus, music theory provides the theoretical framework and conceptual basis for many MIR aspects, including analysis, feature extraction, retrieval, transcription, and generation of music.

• Ethnomusicology. Examining its influence on identity, community, and society, ethnomusicology explores the cultural, social, and historical dimensions of music (see, e.g., Rao et al., 2023; Duan et al., 2023; Tzanetakis, 2014). Drawing insights from this field, MIR can aid in tasks such as collecting and annotating music collections, developing inclusive and culturally diverse music representations, and facilitating cross-cultural comparisons of musical practices and traditions.

• Psychology. The psychology of music explores how music affects human cognition, emotion, perception, and behaviour (Deutschm, 2013). It encompasses areas such as music cognition (Honing, 2021), psychoacoustics (Fastl and Zwicker, 2007), user experience, and music therapy (Agres et al., 2021).

• Acoustics and Audio Engineering. Acoustics is the study of sound and its transmission, while audio engineering focuses on the recording, processing, and reproduction of sound. These fields are essential for understanding the physical properties of musical sound and for developing audio technologies and equipment (see Fletcher and Rossing, 1998, for an example reference in this field).

• Computer Science. Serving as the technical cornerstone and providing essential tools for the efficient and effective analysis, organization, and retrieval of music information, computer science stands as a core discipline within MIR. Key subfields include machine learning, pattern analysis, information retrieval, data representation, and algorithm development.

• Engineering. Techniques particularly in the realm of digital signal processing (DSP) play a crucial role in the analysis, synthesis, and manipulation of music recordings. Leveraging methods such as the Fourier transform, filter design, and time–frequency decompositions, DSP facilitates tasks like spectral analysis, feature extraction, filtering, noise reduction, and audio effects processing (see, e.g., McFee, 2023; Zölzer, 2002).

• Music Technology. This field involves the design and implementation of hardware and software tools for creating, performing, and recording music. It encompasses areas such as electronic music instruments, digital audio workstations (DAWs), music software applications, and interactive music systems.

• Music Education. Music education involves teaching the theory, performance, history, and technology of music across various cultural and academic contexts. MIR can learn from music education by incorporating pedagogical strategies and user experience insights to develop tools that are pedagogically relevant and easy to use. Meanwhile, music education can benefit from these advanced technologies to enhance virtual digital curricula and other educational materials.

Overall, the interdisciplinary nature of music reflects its multifaceted role in human culture, society, and technology, enabling us to draw on insights and methodologies from diverse fields to advance our understanding and appreciation of music.

3.3 Data Multimodality and Richness

Music, being a highly multimodal and rich multimedia domain, engages various modalities including auditory, visual, tactile, emotional, and cultural aspects (see, e.g., Li et al., 2019; Essid and Richard, 2012; Müller et al., 2012). These modalities are closely related to MIR in the following ways:

• Auditory Modality. Music primarily engages the auditory sense, conveying information through sound waves, including pitch, rhythm, melody, harmony, and timbre.

• Visual Modality. Music data often includes visual elements such as musical scores, sheet music, graphical representations of sound (e.g., spectrograms), and music videos. Visual elements provide additional information about the structure, dynamics, and performance of music and vice versa.

• Tactile Modality. Musicians and listeners often experience tactile sensations while playing instruments, feeling vibrations, or tapping along with the rhythm. Additionally, tactile feedback can be incorporated into music interfaces or haptic devices for immersive musical experiences.⁴

• Emotional Modality. Music evokes emotions and expresses feelings through its tonal and rhythmic elements. Although emotional responses to music can vary widely among individuals and cultures, music remains a powerful medium for conveying and eliciting emotions (Cañón et al., 2021; Yang and Chen, 2011).

• Cultural Modality. Music reflects cultural aspects of society, including language, traditions, rituals, and social norms. Different musical styles, genres, and practices emerge from diverse cultural backgrounds, contributing to the richness and diversity of the musical experience.

3.4 Real-World Applications

In recent years, there has been a significant increase in commercial breakthroughs through MIR technology. These range from early examples like audio fingerprinting (Wang, 2003), as used by Shazam, to a wide range of MIR technologies employed by streaming services and other digital service providers. Real-world applications are instrumental in advancing the field of MIR by offering opportunities for validation, feedback, innovation, and collaboration, thus driving research and education by addressing practical challenges. In the following, we outline some key application areas without attempting to be exhaustive.

• Music Recommendation Systems. MIR techniques are used to develop music recommendation systems that suggest personalized playlists, albums, or tracks to users based on their listening history, preferences, and context (Knees and Schedl, 2016). Originally explored within the realm of digital music libraries, these systems have now gained broad usage across streaming platforms, online music stores, and personalized radio services, enhancing user satisfaction and engagement.

• Music Classification and Tagging. MIR algorithms are employed for automatically classifying and tagging music according to genre, mood, tempo, instrumentation, and other musical attributes (Nam et al., 2019). This facilitates music organization, search, and discovery, enabling users to find relevant content more efficiently.

• Music Transcription and Score Following. MIR research addresses the challenges of automatically transcribing audio recordings into symbolic representations, such as musical scores or MIDI files (Benetos et al., 2019). Score following techniques enable computers to synchronize with live performances and generate real-time annotations, facilitating applications in music education, performance analysis, and accompaniment systems (Dannenberg and Raphael, 2006).

• Music Analysis and Visualization. MIR methods enable the analysis and visualization of musical content in various forms, including audio waveforms, spectrograms, chord sequences, and structural annotations. These tools aid music scholars, composers, and performers in studying music structure, style, and expression, as well as in creating visualizations for educational and artistic purposes (Goto and Dannenberg, 2019).

• Music Generation and Composition. MIR techniques contribute to the development of algorithms and systems for generating new musical compositions, improvisations, or arrangements. These systems can be used for automated music composition, collaborative music-making, and creative exploration, inspiring new artistic expressions and fostering musical innovation.

• Music Copyright and Intellectual Property Management. MIR methods play a role in copyright enforcement, content identification, and intellectual property management in the music industry (Yesiler et al., 2021). Techniques such as audio fingerprinting and watermarking are used to detect unauthorized use of copyrighted music and protect the rights of content creators and distributors.

• Digital Humanities. The applications of MIR in digital humanities are multifaceted and significant. For example, MIR technology is increasingly used for the preservation of music-related cultural heritage and facilitates the analysis and comprehension of musical compositions and performances (see, e.g., van Kranenburg et al., 2019; de Valk et al., 2017; Rosenzweig et al., 2022). Furthermore, MIR techniques are utilized in corpus-driven research to analyze extensive music collections, unveiling patterns, genres, styles, and temporal trends (see, e.g., Volk and de Haas, 2013; Weiß et al., 2019; Arthur, 2021; Henry et al., 2024).

• Music Therapy and Healthcare. MIR research contributes to applications in music therapy and healthcare, where music is used as a therapeutic intervention for physical, emotional, and cognitive rehabilitation (Agres et al., 2021). Automated music analysis and generation tools support the development of personalized music therapy interventions tailored to individual needs and preferences.

4.1 Teaching Signal Processing Through Music

As outlined by Müller et al. (2021), music signals offer abundant opportunities to enrich signal processing education, as they evoke intuitive understanding of signal properties that beginners can readily grasp without needing extensive technical knowledge. For instance, on very short time scales, one may want to know the onset position and the fundamental frequency of a played note. On longer time scales, questions may arise about the instrumentation and the chord being played. On even longer time scales, it becomes possible to explore higher-level concepts such as melody and rhythm. Therefore, music can serve as a means to make signal processing education more accessible by employing easily understandable initial questions, which encourage hands-on exploration and experimentation right from the beginning. In particular, music can provide tangible examples to demystify mathematically intricate concepts essential for comprehending advanced signal processing concepts.

In a typical introductory course on digital signal processing, a broad range of topics is covered, including sampling theory, Fourier analysis, convolution and filtering, and time–frequency representations, just to name a few. In their educational article, Müller et al. (2021) illustrate how music enhances signal processing education by providing practical applications and interactive learning experiences, using Fourier analysis as an example. As shown in Figure 1, musical signals provide intuitive examples that demonstrate the effectiveness of short-time Fourier analysis, unveiling noticeable changes in pitch, volume, and other musically significant attributes in spectrogram images. Similarly, there is value in educational materials that aid in comprehending and teaching more advanced signal processing concepts, such as the widely adopted constant-Q transform (CQT) introduced by Brown (1991), its harmonic extension (HCQT) by Bittner et al. (2017), or feature representations derived from deep learning (Kim et al., 2020).

Figure 1

One of the most challenging and fascinating research problems in MIR is automatic music transcription, which seeks to convert an acoustic music signal into a form of musical notation using computational methods. As detailed in the educational article by Benetos et al. (2019), this task encompasses multiple subtasks, including multipitch estimation, onset and offset detection, instrument recognition, beat and rhythm tracking, interpretation of expressive timing and dynamics, and score typesetting. The diverse aspects of music recordings and their acoustic and musical properties provide abundant opportunities to enhance signal processing education. Integrating music into the curriculum makes it more accessible, engaging, and relevant for students with a musical background, thereby enriching their learning journey and deepening their comprehension of the subject.

In addition to providing motivation through tangible music-based scenarios, the availability of well-designed software packages that enhance accessibility to signal processing is indispensable for facilitating interactive learning (Guzdial, 2013). The MIR community has contributed several excellent toolboxes that provide modular source code for processing and analyzing music signals. Notable examples include essentia (Bogdanov et al., 2013), madmom (Böck et al., 2016), Marsyas (Tzanetakis, 2009), Sonic Visualiser (Cannam et al., 2006) and the MIRtoolbox (Lartillot and Toiviainen, 2007). While these toolboxes are mainly designed for research-oriented access to audio processing, the Python package librosa (McFee et al., 2015) and the FMP notebooks (Müller and Zalkow, 2019), alongside its Python package libfmp (Müller and Zalkow, 2021), adopt an explicit educational stance. These toolboxes provide interactive elements to a typical lecture-based signal processing classroom, allowing students to solidify their abilities to apply signal processing concepts in musical examples, thus bridging the gap between education and cutting-edge research.

4.2 Teaching Music Theory Using Computation

The Open Music Theory textbook (OMT)⁵ is a natively-online open educational resource intended to serve as the primary text and workbook for undergraduate music theory curricula. As such, it explicitly targets individuals learning music theory, primarily undergraduates. That said, certain sections of the book utilize computational methods to enrich the study experience, potentially inspiring prospective contributors to the TISMIR education track.

For example, the textbook includes a Harmony Anthology⁶ chapter. Traditional anthologies may be most familiar in the context of literature as a collection of works that are ‘important’ or ‘typical’ of a time and/or place (e.g., American poetry after the Civil War). Similar resources exist in music theory teaching, bringing together examples from a particular historical period and/or for a pedagogical purpose. For further discussion and many references, see Gotham (2019).

OMT’s Harmony Anthology aims to offer an updated version of this resource type for the modern age, with two main differences. First, it operates on a much larger scale. Traditional anthologies typically provide two or three examples due to practical limitations in printing, binding, and selling. In contrast, the OMT Harmony Anthology gathers hundreds of examples, organized around specific chords such as the Neapolitan Sixth. Second, and relatedly, this larger scale shifts the focus away from a few ‘exemplars’ toward a broader collection, encouraging free exploration.

Exploring this collection underscores the subjective nature of analysis. Therefore, individual readers and analysts (including students) should anticipate disagreements and benefit from open debate. Readers can form their own opinions about whether a specific instance ‘counts’ as an example of the category. This effective method of building meta-cognition is not available in resources that only use clear-cut, unambiguous examples (see also Gotham, 2021).

Under the hood, the Harmony Anthology leverages a great deal of MIR work, both in the preparation of the data and in the extent of links serving that goal of free exploration. First, all the scores in the anthology (and many throughout the textbook) result specifically from open source dataset creation efforts that explicitly aim to service both MIR research and public-facing resources (as reported, for example, by Gotham and Jonas, 2022)). The fact that these scores are released under a CC0 (Creative Commons Zero) license, allowing anyone to use, adapt, and distribute the work without restrictions, is highly significant in terms of the FAIR principles discussed above. Second, the analyses are likewise the result of MIR dataset curation, this time from the meta-corpus ‘When in Rome’ (Gotham et al., 2023). Third, the anthology hosts numerous external links, to the score online (for immediate playback online), score on the Github repository (for download without any need for login), and to the extracted example as a small image, which can be browsed with a pop-up window (no re-direct or download required).

Finally, MIR scholars know very well that, even putting aside the subjectivity involved in an analyst making a commitment to a specific chord, the definition and categorization of these chords for anthology purposes bring a whole additional layer of ambiguity. For example, the term ‘modal mixture’ has been core to American music theory teaching for some time now but never received a robust definition until Gotham (2023). This anthology explicitly implements that algorithm and sheds light on the criteria for selection—again, this is an effective way to build students’ meta-cognition.

At this stage, the connection to current research risks becoming too strong and distant from the educational criteria set out above. On the other hand, the absence of a robust definition in existing educational literature demonstrates the need for something, and the openness of the approach (complete with open-source code for the implementation available back at ‘When in Rome’) provides the kind of FAIR principles on which we place such a high premium.

4.3 Teaching Mathematics and Computer Science Through Music

‘Music and mathematics’ is perhaps the oldest form of an explicit connection between music and other disciplines. This connection traces its origins back through the medieval quadrivium, a four-part curriculum comprising arithmetic, geometry, music, and astronomy, all the way to antiquity. Moreover, it continues to be revived and modernized. Noteworthy for our present purposes is the special issue on ‘Pedagogies of Mathematical Music Theory’ in the Journal of Mathematics and Music, where the concept of teaching mathematics through music is explicitly discussed (Yust and Fiore, 2014).

Computer science, despite its shorter history, has also found value in leveraging the enduring connection between mathematics and music in its own distinct ways. Notable examples of using music to instruct basic coding include SonicPi⁷: a code library and GUI for live-coding music. SonicPi explicitly prioritizes the teaching aspect, influencing every decision, down to the naming of attributes and methods. For an example of how music may help teach basic coding, consider the ‘for-loop’. While text and visual representations can be effective, arguably nothing captivates as much as musical data. This is particularly true considering our inherent affinity for repetition in music (Margulis, 2014).

A strong connection between MIR and computer science is evident with the availability of large datasets and computational resources. In this context, machine learning with deep architectures stands out as an advanced approach within MIR (see, e.g. Humphrey et al., 2013; Pons et al., 2018). In particular, the ‘end-to-end’ pipeline represents a significant departure from the established approach of engineering features and then training models. Instead, it integrates these two steps, optimizing features and models together, resulting in MIR systems that seem to perform very well. However, a considerable challenge emerges when it comes to evaluating the resulting ‘black box’ in meaningful ways. This task involves dissecting the system and striving to interpret the functions of its components. An educational article in TISMIR could offer valuable insights by presenting and discussing methods to uncover the workings of the black box.

Expanding on this, while much research focuses on explainable AI, particularly in fields involving text and image data, the music domain offers a unique perspective due to its complexity and multimodality, along with potential confounding factors arising from various versions of the same piece and other inherent intricacies in musical data. For instance, Weiß et al. (2020) explore local key estimation in classical music, using a cross-version dataset of Schubert’s Winterreise performances. Their study reveals insights into model efficacy influenced by different training–test splits and uncovers correlations between errors and annotator disagreements, as well as musically explainable relationships among different keys. As another example, Cosme-Clifford et al. (2023) explore the behavior and constitution of a transformer model, drawing on significant expertise in music theory. This study highlights how music can enhance comprehension in computational domains, providing insights that traditional methods may overlook.

4.4 Teaching Cross-Cultural Comparison Through Computation

Cross-cultural music studies play a significant role in broader cross-cultural education, facilitating the appreciation of cultures beyond one’s own. Given the variations in language, behavior, societal norms, and especially teaching methods, delving into the music of a new culture can pose significant challenges. This is where computational methods prove invaluable, acting as a bridge between different cultural artistic forms. By identifying the commonalities that underlie all music traditions, likely stemming from universal human cognitive processes, it becomes possible to establish connections that greatly enhance understanding of culturally distinctive genres. Examples of this approach are readily found in computational musicology, which not only makes genre-specific music concepts more universally accessible through objective descriptions of audio features and visualization but also benefits from human annotations to provide additional context and insight.

For instance, Clayton et al. (2023) provide examples of vocal performance in the north Indian Dhrupad style, highlighting its meticulously arranged patterns of interaction between syllabic singing and accompanying drumming. They illustrate how the lyrics of the composition are sung at different speeds (basic, double, triple, and quadruple) across various metric cycles with visual depictions of this phenomenon via the derived pitch contour that exhibits shape compression with precisely the same ratios. Tempograms computed on source-separated audio signals allow for the identification of distinct rhythmic episodes throughout the concert, where the singer and the drummer each subdivide the beat differently but in recognizable ratios associated with the intrinsic musical concept of laykari (rhythmic play). This analysis offers insight into the various ways in which a listener familiar with the cultural context may parse and appreciate the music.

Culture-specific attributes of music, such as the precise intonation of scale intervals, can be effectively demonstrated through computational analyses like continuous pitch contours, with platforms like Music in Motion⁸ offering valuable educational resources. Future efforts should focus on simplifying and adapting visualizations generated by MIR to better meet the educational needs of the non-MIR music community. Another important objective is to create educational materials and interfaces to assist musicologists in leveraging MIR to complement traditional approaches to music study. Lastly, as mentioned in Section 4.2, MIR methods have the potential to enhance music pedagogy across cultures by contributing to the development of robust definitions of music concepts. This can play a significant role in cross-cultural educational contexts and foster further research in music across different cultures.