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…
The next part of our exploration into Heaps Curves is to analyze the Skiptune corpus after it has been ordered by year of composition and publication. As this is a blog, we are constantly updating and adding to the database between the weekly publications. At this point, we happen to have an even 83,400 number…
We now turn our attention to Heaps’ Law when our musical word is defined not as a single note, but as a note followed by the next note in a melody. This 2-note pairing will increase the vocabulary size, so we would expect results placing the Skiptune database a little closer to Heaps’ Curves as…
This week we check to see how much of our two choices of ordering the database, as entered and by year, are typical of random orderings. We test that by shuffling the order of the database 1,000 times, and then running the Heaps Law analysis one thousand times, one for each row or tune in…
Last week we found that Heaps’ Law applies well to the Skiptune corpus of melodies with a beta value of 0.36 using a note as a “word”. But as we observed, the order in which a computer reads in the database can affect the curve fit, so this week we check that out to see…