Challenging epistemic biases in musical AI: a guerrilla approach to human – machine comprovisation based on Xenakis’s sketches for Evryali

Article by Pavlos Antoniadis, Department of Music Studies, University of Ioannina.

Download article

Abstract: The objective of this paper is to reflect on the affordances of sketches as interfaces for human and machine learning, by way of a case-study based on Iannis Xenakis’s Evryali (1973). First, we report on one-to-one mappings between the composer’s original sketches and the symbolic notation intended for performance. Then, we outline the sketches’ deviations from the symbolic score and their potential in offering indispensable analytical insights for learning. The decoupling sketch-score intensifies as performance multimodal data enter the framework of our analysis, allowing for the emergence of one-to-many mappings among those three distinct representation domains. This multiplicity of relations fuels the creation of gesture-controlled, augmented, and interactive tablatures, which are based on the sketches and incorporate graphic and multimodal elements to bypass conventional notation.Finally, we report on the use of tablatures as both preparation and performance tools in a human – machine comprovisation setting, involving a human trained to improvise on complex scores, an AI agent trained on a corpus of recordings, and a gesture – follower trained on the performance ofsketches. As a postlude, we point to the potential of one-to-many mappings for challenging established epistemic biases in musical AI. We capitalize on the unpredictability generated by the interplay between couplings and decouplings of different representation domains, affirming the transitory nature and inherent malleability of sketches.

Laisser un commentaire

Votre adresse e-mail ne sera pas publiée. Les champs obligatoires sont indiqués avec *