I’ve been working for a while on an app that makes up melodies. Each melody is composed of several phrases (musical sentences, or ‘sub-melodies’).
Now, I want to improve on this:
- The user can press a button to ‘like’ a melody.
- The app saves the phrases (sequence of notes) composing the melody to storage.
- These stored phrases will be integrated into future melodies.
In particular, the app should not only use the verbatim phrases but apply attributes like the rhythm, pitch etc. to generate new melodies.
My question is: Which machine learning techniques can I apply to infer new melodies using the stored phrases?
5
Sure it is. It’s supervised machine learning, in that you use preexisting external expert judgments to try to deduce the essence of a pattern (in this case, “pretty melodies”) and ultimately allow you to generate more of them autonomously. The fact that the metric you are optimizing (aesthetic judgment on music) is somewhat unusual doesn’t change that.