Personalizing AI for Co-Creative Music Composition from Melody to Structure - Archive ouverte HAL Access content directly
Conference Papers Year : 2022

Personalizing AI for Co-Creative Music Composition from Melody to Structure

Ken Déguernel
Richard Groult
Sebastien Gulluni
  • Function : Author

Abstract

Co-creativity is a unique artistic situation where human interact with computer, and raises challenges on interaction, steerability, and personalization. We present a new co-creative music composition approach that we used for our participation in the "AI Song Contest 2021", an international music contest involving artificial intelligence (AI). We *personalize* the artificial creativity methods to adapt to the needs and expectations of a composer. Interactions between the composer and different AI methods occurred throughout the whole composition process, for the generation of melodies, chord progressions, global structure, and textural variations, both through *data sharing* for machine learning based AI and through *knowledge sharing* for rule-based AI. We describe these AI methods and how the composer interacted with them: The personalization of the AI enabled the composer to explore new musical territories while keeping their original style, with AI music generation which "sounds like it has been generated for him''. The song "The last moment before you fly" ranked 3rd place in this contest, the jury underlining the "personal feel" of the song. We discuss here how these methods open the way to new co-creative approaches, both using AI and personalization.
Fichier principal
Vignette du fichier
2022-smc-personalizing-ai.pdf (627.64 Ko) Télécharger le fichier
Origin Publisher files allowed on an open archive

Dates and versions

hal-03618015 , version 1 (07-06-2022)

Identifiers

Cite

Ken Déguernel, Mathieu Giraud, Richard Groult, Sebastien Gulluni. Personalizing AI for Co-Creative Music Composition from Melody to Structure. Sound and Music Computing (SMC 2022), 2022, Saint-Étienne, France. pp.314-321, ⟨10.5281/zenodo.6573287⟩. ⟨hal-03618015⟩
385 View
504 Download

Altmetric

Share

Gmail Mastodon Facebook X LinkedIn More