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.