In a society in which digital technology has become a critical component, it becomes increasingly important to be aware of ways in which the integration of technology may have consequences to the way in which societal challenges are perceived and understood. In this talk, I will reflect on this, wearing my different hats as a computer scientist, educator, musician, inter- and transdisciplinary researcher, and a recent public figure in AI policy and media commentaries. Through multiple examples, ranging from MIR and the cultural sector to recent cases of algorithmic discrimination, I will illustrate methodological and conceptual points of friction between what a technologist may usually optimize for, and what may be needed in practice. Based on this, I will emphasize the explicit need for a holistic view on our professional responsibility as digital experts.

Indian music is renowned for its rich conceptual frameworks, diverse compositional forms, aesthetic depth, and inherent logical structures. Exploring how abstract concepts can serve as catalysts for musical creation presents a compelling and prospective area of research. This lecture–demonstration examines the process through which conceptual ideas evolve into innovative musical compositions and compositional forms.

The presentation features a selection of original compositions by the researcher, each derived from distinct conceptual foundations. The lecture will illustrate several approaches, including the translation of mathematical concepts into compositional structures, the extraction and reorganization of rhythmic patterns as the basis for new works, and the inter-transfer of conceptual ideas between different musical systems.

Additionally, the lecture highlights a unique and original concept conceived by the researcher’s teacher (guru), where an exclusive principle drawn from language has been transformed into a new compositional form.

Through analytical discussion and live musical demonstrations, the lecture aims to reveal how interdisciplinary and conceptual thinking can expand the creative and structural possibilities within Indian music.

Pending

Pending

Many generative AI (GenAI) tools can produce high quality content from simple text prompts. For many creative practitioners this makes such tools unusable and uninteresting for their creative practice. Instead there is an opportunity to explore GenAI tools which better support the praxis of intangible cultural heritage (ICH) rather than attempting to create the most accurate or complete GenAI renderings of ICH. I refer to this as ICH praxis oriented GenAI, or praxis GenAI for short. I purposefully place emphasis on the praxis of ICH before GenAI to foreground that in this approach ICH is more important than GenAI, not the other way around as is often the case in technocentric approaches to ICH. Praxis GenAI relies on a pragmatic balance between ICH lived practice and creative skills with the generative accuracy (or not) of GenAI. Praxis GenAI’s role then is the generation of content that is close enough to a practitioner’s style and yet flexible enough to be useful in their creative practice. Thinking of praxis GenAI as supporting and responding to lived practice in this way also offers pathways to explore, probe, and better understand ICH rather than seeing GenAI as leading ICH innovation. I illustrate my talk with examples from current research including my own research on music and embroidery with GenAI, highlighting how small, lightweight GenAI models offer opportunities for creative expression and reflection, creative agency, and the responsible use of AI.

In this talk I will present the work by the Music and Sound Cultures Research Group in recording and digitally encoding music from the Arabian Gulf. This music is interestingly rich in influences that span from East Africa to South India, yet it is not well known outside this region. I will describe the challenges posed to researchers interested in researching this music, and discuss the transdisciplinary methodologies we are developing in order to make this music discoverable and analyzable in the digital domain.

In this presentation, I will share some of my personal reflections and reactions on the widespread and often polarised discussions surrouding the recent boom of generative AI in music, both in mainstream conversations and within the MIR community. I will first position this incursion within the broader context of Computational Creativity and regarding the musicological position about technology and industry. From there I will discuss these themes through the lense of the emerging AI Music Studies field, in particular by drawing on insights and debates that emerged during a recent workshop on the Ethics of AI in Music that took place in KTH, Stockholm, last November.

Gregorian chant was a central musical tradition in Medieval Latin Europe and one of the expressions of its cultural unity: any Latin Christian liturgy, such as the weekly Sunday mass, would have involved prescribed chants as a major part of the ritual. The practitioners were required to conserve chant melodies; nevertheless, a single chant rarely has the exact same melody in any two surviving manuscripts, resembling folksong more than ”works” traditionally studied by historical musicology. Some systematic patterns within this melodic diversity have previously been observed in chant scholarship, but the scale of chant makes this study difficult without automation. We notice remarkable analogies between biological evolution and processes of Gregorian chant transmission, which lead us to suggest recovering these “melodic dialects” using phylogenetic methods. We show that phylogenetic models recover historically plausible patterns of chant melody evolution. Some, but not all, institutional networks play a more important role than geographic proximity. Additionally, extreme phenomena in the inferred phylogenies draw attention to exceptional historical circumstance. Beyond Gregorian chant itself, this framing of melodic diversity as a result of transmission processes becomes a pathway towards mapping the underlying ecclesiastical cultural networks.

In the South Indian music tradition known as Carnatic music, svaras (notes) are often performed with gamakas (ornamentation), which can involve oscillations, slides and brief movements to neighbouring pitches. Musicians and musicologists have long observed that the choice of gamaka is dependent on melodic context, a phenomenon recently theorized using the concept of coarticulation: where the performance of a unit (in this case a svara) is influenced by that which precedes or follows it. In this talk, we discuss and analyse the coarticulatory nature of Carnatic music and demonstrate how incorporating these processes into our models can lead to improved svara representation learning.

While research is sometimes a solitary pursuit, most recognise that deep, field-wide collaboration is the best way to advance the field most effectively. And while MIR – like all fields – faces certain challenges, overall, it is a wonderfully open and engaged community that is highly responsive to this best practice and keen to build each other up for the collective good. This is a key strength to build on. In this talk, I will share frank experiences from some major, long-term and highly collaborative projects including the “OpenScore" transcription effort (in collaboration with the commercial company Musescore.com and a large number of community volunteers), the "Open Music Theory” textbook (http://viva.pressbooks.pub/openmusictheory/), the social initiative "Four Score and More" (https://fourscoreandmore.org/), and a nascent effort to coordinate a wide range of Algorithms for Music Analysis and Data Science ("AMADS, https://github.com/music-computing/amads/).

This work presents an attempt to study the aesthetics of North Indian Khayal music with reference to the flexibility exercised by artists in performing popular compositions. We study expressive timing and pitch variations of the given lyrical content within and across performances and propose computational representations that can discriminate between different performances of the same song in terms of expression. We present the necessary audio processing and annotation procedures, and discuss our observations and insights from the analysis of a dataset of two songs in two ragas each rendered by ten prominent artists.

Far from being a solved problem, deep learning source separation has reached impressive performance, particularly on Western pop-rock music. Formalising the task as a 4-stem separation problem—alongside open baselines, datasets, and public challenges—has driven significant methodological progress. However, the task remains challenging in non-standard setups, such as variable instrumentations, non-Western genres, universal source separation, and bioacoustics, where clean reference data is often unavailable. This talk will explore several strategies to address this issue, including domain adaptation, self-training, and knowledge distillation, and will present a practical case: denoising animal vocalisations.