Towards Co-Evolutionary Musical Improvisation

Marco Scirea, Julian Togelius, Peter Eklund, Sebastian Risi

Paradigms in Algorithmic Music


MetaCompose, Scirea et al.

Warning: Shameless self-publicity

Paradigms in Algorithmic Music


Paradigms in Algorithmic Music

Live Algorithms:
live music performance systems that, to some extent, exhibit autonomy or agency


  • Modular
  • Collaborative with a human
  • Adaptive - through coevolution
  • Autonomous music production

Rhythm input/output


NeuroEvolution of Augmenting Topologies (NEAT) with collaborative evolution: n-tuples of ANNs are evaluated together




Pitch fitness

Rhythm fitness


Both fitnesses

Research Question

Can the system create emergent patterns with very simple fitness functions?

  • The emergent music patterns are not just caused by the architecture of the system
  • The introduction of small evolutionary pressure is enough to perceptively improve the result

Experiment Design

Quantitative study, participants asked to pick a preference between two clips on pleasantness, randomness, harmoniousness and interestingness

538 answers from 87 participants.

Music clip generation

8 melodies were chosen from a range of different genres

2 clips were evolved (one using a random fitness and one with a defined fitness) for each melody (100 generations)

Total of 16 clips.

All pieces can be found at



Randomness is statistically significant, but much less than the others.
  • High freedom leads to uncommon music
  • Small lenght of pieces might make it hard to discern emergent patters

In short

  • The introduction of a simple fitness function perceptively improves the output of the system
  • Randomness seems to be a harder to distinguish between the pieces


The system is still in it's infancy

Too many modules = cacophony

Can more complex fitnesses be defined, leading to more complex music?

What now?

Qualitative experiments with musicians playing with the system in real-time, in order to find out how the interaction between the players unfolds, and what kind of creative improvisations can emerge.

Thank you :)

Webpage: Twitter: @MarcoScirea