The award-winning 'Hamiltonian Cat Flap' algorithmic music system

HCF is a computer-aided composition system I built in 2021. It uses AI constraint programming techniques, graph theory and automata theory. It was built using Python, Python libraries such as Graphviz, Matplotlib and Mido, and the Lilypond tool.

In late 2022, when I had mostly forgotten about it all, I was dead chuffed to be told that HCF had won the 2020/2021 UoL Prize for Best Project.

Musical Turing Test

I generated some natural and realistic sounding melodies with HCF, using melody corpora built from varying selections of British and Irish folk songs. I then ran a 'Musical Turing Test' experiment using human listeners to see if they could distinguish real folk song melodies from my computer generated melodies. The results of the MTT were a little surprising as I personally considered the 3 melodies to be in increasing order of realism, which is the opposite order to how listeners performed in the listening tests. However, the human listener sample size was small, all 3 of the melodies managed to deceive a number of listeners, and overall human listeners were deceived in 46% of listening tests, so the results were still pleasing. The first melody in the 4 melody MTT was a deliberately unrealistic 'obvious fake' test of listener musicality and is not included here.

The 3 melodies were presented in a very basic and raw form, being only simple MIDI piano melodies and accompanied by only simple MIDI percussion tracks to indicate the appropriate beat stresses. These should not be thought of as proper musical recordings for listening enjoyment, but rather as representative examples of the melodies that HCF can produce.


MTT Test B - 69% of listeners mistakenly thought that this was a real melody:




MTT Test C - 38% of listeners mistakenly thought that this was a real melody:




MTT Test D - 31% of listeners mistakenly thought that this was a real melody: