Grow

What's going on here

I sang a short phrase, and on Day 3 a program turned it into notes. Here, simple rules change those notes: move them up, turn the tune upside down, add notes, push the rhythm about. Each card shows the result of one rule, and one card combines two. There's no synth: every variation is my recording of the phrase, re-pitched note by note.

The number on each card says how closely the variation keeps the shape of the tune and the rhythm of the original, where 1 means the same. Closer cards sit nearer the middle. The small print names the rule and how strongly it was applied.

What kept the family

By that number, nearly all of them stay close, at 0.89 or above. Only turning the tune upside down moves far away, at 0.67. The number has limits: it doesn't notice how high the phrase sits or how loud it is, so moving it up a fourth and changing only its loudness both score 1.00. It can't say whether they still sound like family. That takes listening.

Where this is heading

Days 3, 4 and 5 are steps towards an overnight composition factory: I record a phrase on my phone, and by morning it has become a set of sketches to listen to. This step could run as ordinary code, because the same phrase and the same rules always give the same variations, so any sketch can be traced back to what made it. I'd keep the sketches I like, and the rules behind them would count for more next time. An agent could handle that part, looking at what I kept and suggesting new rule strengths for me to accept or not.

The rules and the notes they produced are in family.json.