ARTFEED — Contemporary Art Intelligence

MACAT and MACataRT: New AI Systems for Co-Creative Music Performance

ai-technology · 2026-08-17

Two innovative musical agent systems, MACAT and MACataRT, have been unveiled by researchers to enhance collaborative music performance and improvisation between humans and AI. MACAT focuses on agent-led performances, utilizing real-time synthesis and self-listening to independently refine its output. In contrast, MACataRT provides a versatile setting for joint improvisation through audio mosaicing and sequence-based learning. Both systems are trained on tailored, small datasets, promoting ethical AI engagement while maintaining artistic integrity. This research, detailed in a paper on arXiv (arXiv:2502.00023v2), emphasizes the role of interactive, artist-centered generative AI in broadening creative avenues for musicians in live performance. It advocates for a transition toward AI tools that prioritize user control and ethical practices in the arts.

Key facts

  • Two musical agent systems, MACAT and MACataRT, have been developed for co-creative music performance.
  • MACAT is optimized for agent-led performance, using real-time synthesis and self-listening.
  • MACataRT provides a flexible environment for collaborative improvisation through audio mosaicing and sequence-based learning.
  • Both systems train on personalized, small datasets to ensure ethical and transparent AI engagement.
  • The research emphasizes artist-centred generative AI to expand creative possibilities in real-time performance.
  • The paper is available on arXiv with identifier 2502.00023v2.
  • The announcement type is replace-cross, indicating a revised version.
  • The systems are designed as human-in-the-loop generative AI for music.

Entities

Institutions

  • arXiv

Sources