This blog documents the process of teaching an AI system to generate melodies from the Skiptune database. Expect experiments, mistakes, and occasional surprises, and at the end expect melodies that sound as if a human wrote them.
Problem to Be Solved The Skiptune database is a database of 83,000 melodies from around the world and across several centuries. These melodies are encoded in a special way (a pitch difference followed by a duration ratio), and we want to explore whether an AI model can learn melodic structure from this representation. The…
This week we lay out our plan for figuring out the search paradox, which is that tunes have a widely variety of overlap with their two-note patterns (tuples), from zero to thousands, sometimes contrary to how common the notes look to the eye. We have to control for genre. The ratio of common to rare…
Our workflow includes a step to make sure the tune we’re entering into the database is not already there. For that we devised a search engine that, as a by-product, tells us how many tunes already in the database have all the search terms in them. In doing so over the years, we noticed something…
There are many models to choose from when training a musical corpus. We’ll run through the main candidates briefly and explain why we settled on the transformer model with attention when we turn to AI training in the future. Artificial Intelligence Models AI models were designed for various purposes, such as pictures, words, and classifying…
Having dealt with both the distribution of durations, duration ratios, and rests in the Skiptune database, the last item is the pitch differential: The difference between any note’s pitch and its neighbor on either side. Let’s start with the basic counts: 4,446,196 note-to-note transitions 92.7 percent of all transitions are between notes 7.3 percent involve…