Just because this section is fairly quiet, I'll spam some old algorithmic composition experiments. It's not world changing stuff, but it explores a few interesting ideas.

Really great sounding stuff! Thanks for sharing! I especially like the 12TET video; I find the that the overall tonal structure in that one is something I can actually follow musically, where the others are a little bit more difficult to follow (but still interesting).

Is all of the synthesis done in PD, too? The quality of your sounds always surprises me; I would expect these kinds of sounds from hardware or pricy soft-synths, but actually developing instruments that sound that good takes quite a bit of dedication and a deep understanding of sound. The mix also sounds very nice.

I feel like even though these kinds of patches are really fascinating, they don't get shared too often on these forums, so thanks again for posting!

It's hard with algorithmic music to find some balance between novel and incomprehensible. I don't mind if it's a little naive or strange sounding, since that's really part of what makes it interesting in the first place, but it takes some effort to make it not sound like a total mess. If I make it too "normal" sounding, it's usually very interesting. I have gotten some better material out of these algorithms (if I ever get around to mixing it), but I think they're not really as flexible as I'd hoped. I have some better ideas...

The sounds and effects are all PD. It's actually mostly pretty simple.

@acreil said:

It's hard with algorithmic music to find some balance between novel and incomprehensible. I don't mind if it's a little naive or strange sounding, since that's really part of what makes it interesting in the first place, but it takes some effort to make it not sound like a total mess.

I find your 'just intonation' piece an excellent example where that balance is achieved. Familiar enough to sound musical and strange enough to attract the attention. Would you mind to reveal some details of 'a weighted selection of integer frequency ratios'? Which ratio's did you use, what means 'weighted selection'? How do you make patterns evolve?

Katja

The original idea was to take some initial note (i.e. 100 Hz) and multiply it by some randomly selected prime factor (1/11, 1/7, 1/5, 1/3, 1/2, 2, 3, 5, 7, 11). Successive iterations can produce about any arbitrary interval, without sticking to any particular scale or absolute frequency. Of course a completely random selection will just go all over the place without any bounds, and you might want the more "exotic" sounding things to occur less often. So I made a normal distribution that would shift left or right according to the current frequency, thus keeping it more or less within a musical range. And I added another thing to prevent it from repeatedly selecting the same factors (i.e. alternating 1/2 and 2 and effectively going nowhere). In some versions I left out the 1/2 and 2 factors entirely, to try to make it do something more interesting.

The resulting frequencies are just round-robin assigned to your tone generator of choice. So the last n notes overlap to form a chord of some sort. The arpeggio part is just some sequence of harmonics of the most recent note.

That sounds fine on paper, but it doesn't really work that well. First, it tends to stick to less interesting territory (the center of the distribution), but the other factors are too extreme (>2 octaves). So the end result is that it's mostly "vanilla" sounding (1/3 and 3) until it selects something in the tail of the distribution. Then it makes an obnoxiously low or high note and corrects with some large factor that clashes with the previous notes. I found it very difficult to get it to behave.

I think it's a mistake to use the normal distribution. I should be using some limited set of ratios (that maybe change over time) and "hard" limits, i.e. move an octave up or down to stay within specific bounds. I'm still thinking about it.

@acreil said:

I found it very difficult to get it to behave.

Yeah it's hard to build a robot-composer which produces an unconventional yet esthetically pleasing output every time. I'm struggling with the same issue, even though I use acoustic input as the source for patterns instead of completely algorithmic works.

I'm now experimenting with a 5-tone equal temperament scale (approximately the slendro scale of Balinese Gamelan). In this scale, only the smallest interval sounds rough. The 5 intervals form a tonal grid rather than a scale. Every tone can serve as a center of gravity. The point of gravity is in the low, long notes, and the higher pitched parts may use all notes in the grid. Slendro makes a very convenient tone grid. With limited material you can do a lot of different impressions. Still, my pattern-generator can't make the most of it yet.

Maybe I could incorporate your concept of differential tones somehow. In most musics, sequences of notes do not jump wildly or even randomly over the scale all the time, but rather tend to meander around temporary anchor notes. A differential system with a normal distribution seems appropriate for that purpose. The problem with the pure harmonic intervals is, as you mentioned, there are too little of them in a reasonable range. So, maybe the concept works better for melodies (where you can use smaller intervals) than for harmonies.

Katja

I'm mostly disappointed with existing algorithmic composition tools and methods. I haven't really researched it that much, but the stuff that isn't knowledge-based (I'm not interested in that) tends to either be dissonant serialist type stuff (which I don't really mind, but it strikes me as sort of obsolete) or just major scale elevator music (fine for several minutes but it quickly becomes boring). I know a lot of computer music goes into microtonal/generative territory, but I mostly find it unsatisfying. I need to get something that's consonant but also explores other interesting territory.

I have some better ideas for equal tempered scales. In 12TET I've mapped out all the useful 7 note scales and transitions between them. It can similarly be done for chords (3 or 4 out of 7). You can make a sort of histogram of intervals and chords for each scale and rank them according to how consonant or dissonant they are, then make choices based on that. I'm trying to work it out in 19TET but there's a huge number of possible scales. And aside from some basic constraints, I'm not familiar enough with it to be sure which ones are actually useful.

I've been thinking about it a little more, and doing some reading... maybe it would be better to consider the 3rd, 5th and 7th harmonics as sort of independent axes. Octaves could be used freely to keep things within a certain frequency range (and maybe to enforce a certain "melodic" quality, i.e. not too close but not too far away), and maybe the probabilities of the individual factors could vary over time (mostly 5 or 1/5 at one point, mostly 3 and 7 at another ) to encourage some sort of coherent flavor.

I also wrote a program to find all the useful scales in 19TET (according to arbitrary constraints), and their relationships. Doing it on paper was getting to be ridiculous. I don't think anyone's done that before. It should be extremely useful.