Jamming with Genetic Algorithms

Notes:

* The genetic algorithm is used to evolve chords.
* I'm a terrible drummer.

* http://mingus.googlecode.com/

Category:

Science & Technology

Tags:

• mingus
• drum
• genetic
• algorithm
• audio
• python
• jam
• nerd
• software

[Onderstekop]

What I'm getting at is that a chord is made up out of more than one frequency, so how would you make that work. Should you still solve for 0 or should you just compare on a per note basis? Maybe it would be easier to have a fitness function that just counts thirds and fifths.

[WhiteDragon103]

but if you graphed out:

y = sin(x) + sin(x*1.4173759), it would take many, many revolutions for both sin functions to equal 0.

[WhiteDragon103]

Well, the point of this kind of genetic algorithm, where fitness is determined by the harmony of the chord, the point would be to have good-sounding chords randomly generated as if the computer was the composer. But I suppose intervention would be possible, as genetic algorithms are flexible.

[WhiteDragon103]

err, i made a mistake:

If you graphed this out, sin(x) AND sin(x*1.5) would both equal zero when X is 0, 3, 6, 12 and so on*

[WhiteDragon103]

I am guessing that because you are a programmer that you know what sin() and cos() do.

Well, imagine this formula:

y = sin(x) + sin(x * 1.5)

If you graphed this out, y would equal 0 when X is 0, 3, 6, 12, and so on (assuming sin(1) meant one whole revolution, like sin(360 degrees) or sin(pi * 2))

[Onderstekop]

It sounds interesting, but how would you get the fitness for for instance D, E, F when you are looking for C, E, G?

[WhiteDragon103]

Imagine frequencies and harmony as something like planets orbiting around the sun. If planet A orbits the sun 1.5 times sun then planet B, it will take planet A 3 revolutions for planet B to make 2 revolutions, and after that point you would be back where you originally were. However, if the ratio was something like 2:5.5123 rather than 2:3, it would take much longer for the planets to meet up again at the same point they originally were.

[WhiteDragon103]

Okay, I see.

However, I have something I want you to try:

Have fitness be determined by the amount of time it would take for the chord to reach a repetition point in the waveform. A harmonious chord is one where the waveform comes to a repetition sooner than a non-harmonious chord, relative to the same note. For example: C sounds harmonious when played with a G, because they are fifths, meaning that mathematically G is 1.5 times the wavelength of C. .....

[Onderstekop]

You are understanding correctly that this is not an algorithm that finds _some_ chord, because a lot of random notes make up some kind of exotic chord so that GA would be done in 1 or 2 generations and thus not so interesting.

I was more interested in the semi chaotic process that you could play against when you know what chord you are looking for. It introduces variations.

It's evolving from a random collection of notes to the one you 'want' and plays the best chord in every generation.

[WhiteDragon103]

Perhaps I am not fully understanding the terms you are using, but if you can use a function to determine the next set of chords, why use a genetic algorithm to find it?