Haebichan Jung
Sep 7, 2018 · 1 min read

Great comment, and certainly one I’ve been thinking about. As a composer myself, I definitely move back and forth between what you call the “creative” and the “rote” processes in musical composition. From a sheer music standpoint, I see that dual interaction captured by the melody and the harmony. If the ‘Axis of Awesome’ proves anything, it’s that the harmony is the rote repetitive part and the melody is that “creative” part of popular music. This makes sense if you think about how jazz improvisation works. The harmony repeats in the same automatic cycle, and the musicians insert their creativity only in the melody. I think that’s why a huge chunk of pop music theory can be gleaned from a narrow sample of data, due to these samples sharing the same phenomenon (my website actually is designed to prove that philosophic idea: if you generate pop music based on 2 samples, would that make a huge musical difference compared to when you generate music based on say 100 samples?).

In regards to your second point, I am planning to build a AI/Mechanical model that you mention. There are scholars who have a frequentist deep learning method hybridized with a Bayesian Hidden Markov Model to generate pop music. I would like to build something similar as well. If you’re interested in these articles, please let me know.

    Haebichan Jung

    Written by

    Project Lead @ TDS | Data Scientist @ Recurly