Introduction to Generative Music

I first discovered Brian Eno when I began enjoying ambient music. Eno is credited with inventing the phrase “ambient music,” first using the term to describe his 1978 release Ambient 1: Music for Airports.

Ambient 1: Music for Airports

I soon learned of the interesting methods Eno used to compose the tracks for the album. For example, track “2/1” was created using several loops of magnetic tape played simultaneously. Each tape loop contained a recording of a single note, and would repeat on an interval determined by the length of the loop.

“One of the notes repeats every 23 1/2 seconds. It is in fact a long loop running around a series of tubular aluminum chairs in Conny Plank’s studio. The next lowest loop repeats every 25 7/8 seconds or something like that. The third one every 29 15/16 seconds or something. What I mean is they all repeat in cycles that are called incommensurable — they are not likely to come back into sync again.” — Brian Eno

Below is a visual representation of this idea, in which three dots move in a circle. Because each dot circles around for a slightly different duration, you’ll never see them all reach the top at exactly the same time.

Incommensurable dots

The recorded track “2/1” on Ambient 1 is but a short sample of the output from a system Eno built which generated the music. A similar system using the three intervals quoted above would not repeat itself for almost 27 days (I’ll spare you the math).

In case it doesn’t skip to it automatically, Track “2/1” starts at 17:20.

All of the tracks on Ambient 1 involved the use of some system to generate music, a technique Eno also used on Discreet Music, Ambient 4: On Land, Thursday Afternoon, Neroli, Reflection, and probably others albums as well. Perhaps it’s not surprising Eno is credited with popularizing (if not coining) the phrase “generative music” to describe music created this way.

Eno’s Definition

Eno included an essay on generative music in his published diary, A Year With Swollen Appendices:

“One of my long-term interests has been the invention of ‘machines’ and ‘systems’ that could produce musical and visual experiences… [T]he point of them was to make music with materials and processes I specified, but in combinations and interactions that I did not.” — From “Generative Music” in A Year With Swollen Appendices

Eno further describes systems which create “ever-changing music,” that could “produce original music forever,” lamenting the limitations which forced him to simply record the output of these systems as traditional albums rather than “sell the system itself, so that a listener would know that the music was always unique.”

Considering Brian Eno’s descriptions of generative music, two criteria for music to qualify as “generative” emerge:

  1. It must change continuously and never repeat itself exactly.
  2. It must last forever.

Satisfying one of these conditions is trivial. Any existing song could simply be played over and over again, and the resulting piece would last forever. Anyone could play random keys on a keyboard for a few minutes without repeating themselves. It’s the combination of these two requirements which specifies something really special: an endless supply of music never heard before.

On “Never Repeating”

When it’s said that generative music never repeats itself, it’s important to clarify what is meant by “repeats.” This isn’t to say generative music can only use a given note or melody once; in this case, a repetition is more abstract than a single note or phrase. Think back to the system which generated “2/1” described above and imagine trying to write down a procedure for playing the piece on a piano. Even if you didn’t understand traditional sheet music, you could write it as a series of steps that would look something like:

  1. Play an E
  2. Wait for 3023 milliseconds
  3. Play a D
  4. Wait for 2044 milliseconds
  5. Play a G
  6. Wait for 1987 milliseconds
  7. Play a C
  8. Wait for 2232 milliseconds
  9. Play a G again

You could carry on like this, writing several thousands steps, and never once be able to write “Go back to step N,” or “Redo steps N through M.” The opportunity to do so wouldn’t present itself (at least not for a very long time; more on that in a moment).

This is different from ordinary music. Consider a chorus. It’s commonplace for a song to feature its chorus multiple times, each instance more or less a copy of the last; that’s how you know it’s the chorus. In fact, it’s not unusual in recorded music for each chorus to be precisely the same as the others, using the exact same recordings of the instrumentals and vocals — as if they were copy and pasted from one part of the song to the others (oftentimes that’s exactly what happens during the mixing process).

A recognizable pattern or section of a song like a chorus won’t be found in a piece of generative music. It would be easy to record some audio and simply loop it to play over and over, but the result wouldn’t be generative music.

On “Lasting Forever”

In the excerpt from Brian Eno’s lecture above, he claimed the timing of the intervals used to generate “2/1” were “incommensurable.” Eno is using the mathematical definition of the word:

(of numbers) in a ratio that cannot be expressed as a ratio of integers

He explains that as a symptom of the tape loops being incommensurable, “they are not likely to come back into sync again.” As mentioned previously, the example timings he gave (23 1/2, 25 7/8, and 29 15/16 seconds) would eventually re-synchronize, but not for almost 27 days.

Contrast the animation of the three dots above with the one below. In this example, one of the dots takes 3 seconds to loop around, another takes 4 seconds, and the third takes 6 seconds. Since all of these numbers are factors of 12 (which is to say that 12 can be divided evenly by any of 3, 4, or 6), the dots will re-synchronize every 12 seconds.

Commensurable dots

It’s easy to see that music created by a system similar to that of “2/1” in which the timings of the notes were all factors of some small number would be rather boring and repetitious, obviously violating the first condition of generative music. But even though the example Eno gave wouldn’t repeat for a long time, it does eventually repeat. Doesn’t that break the rule too?

Technically, yes. But while it’s true that such a system would repeat itself after almost 27 days, no one is actually going to listen for that long. If anyone dared to try, I would defy them to recognize any portion of the music as a repetition.

So while it’s true that technically such a system would not “produce original music forever,” a listener with no knowledge of how the system works would be unable to tell the difference between it and a system which truly generates a never-ending stream of unique music, and that’s good enough. The time it would take a listener to recognize the repetition is far greater than the time said listener will spend actually listening to the piece. Effectively, for any given listener, the music created by the system is unique and original for as long as they hear it. From the listener’s perspective, the difference between “forever” and “as long as I listen” is irrelevant.

A Refined Interpretation

As the systems designer, Brian Eno is surely aware that some of his generative music systems don’t really last “forever” without repetition, but it’s easier to say than, “forever, or at least as near as makes no difference.” While it is possible to design systems which truly generate original music endlessly, it’s not necessary. Here the word “forever” can be interpreted as relative to any single listener’s relationship with the music. With that in mind, modifications to the criteria for generative music are in order:

  1. It must change continuously with no discernible repetition.
  2. It must last as long as anyone is willing to listen.

These rules are less absolute than the previous ones, and are open to some interpretation by creators and listeners alike. But the beauty of these more accurate criteria is they are effectively the same as the old ones; if the listener never notices a repetition nor the potential for one, to them the music may as well never repeat. Likewise, if the system can uphold this illusion as long as one listens to it, then it may as well last forever.

Perhaps there exist some generative music “purists” who demand to have the inner workings of the systems revealed in order to verify whether the music truly is endlessly unique. Without an understanding of a system’s design, there’s really no way to know. It’s a neat trick generative music composers can play on the listener. The art of this illusion is determining the appropriate capability for a system’s output to seem as though it’s endlessly unique. Is 27 days of original music really enough? What about a year? How much time does it take to qualify as “basically forever?”

A Different Type of Music

“From now on there are three alternatives: live music, recorded music and generative music. Generative music enjoys some of the benefits of both its ancestors. Like live music, it is always different. Like recorded music, it is free of time-and-place limitations — you can hear it when you want and where you want.” — From “Generative Music” in A Year With Swollen Appendices by Brian Eno

Prior to the invention of sound recording and playback technology, music was always guaranteed to be unique. Even the same tune played twice would necessarily have differences in each performance. For the vast majority of human and music history, this was simply a reality.

To put this in perspective, humans are believed to be around 200,000 years old. Technology capable of recording audio and playing it back wasn’t invented until 1877, making it just 142 years young. So while “recorded music” may well be synonymous with “music” to the average listener, it’s only been available for less than one-tenth of one percent of the whole of human history.

Generative music offers an opportunity to experience the joy of listening to music that will never be heard quite the same way again, combined with the convenience and portability of the system which generates it. Software-based generative music systems provide the ultimate experience: unique, original, endless music played from a computer or smartphone — the exact same devices used to listen to recorded music today. Brian Eno perfectly captured the exciting potential of generative music in this quote from A Year With Swollen Appendices:

“[I]t’s possible that our grandchildren will look at us in wonder and say, ‘You mean you used to listen to exactly the same thing over and over again?’”

If you like ambient music or are interested in generative music, you might enjoy my own generative music pieces. They can be heard on generative.fm. I find the music is perfect for listening to during activities which require concentration, like reading, writing, or working.
I would be remiss if I did not mention Tero Parviainen’s wonderful, interactive web page, “How Generative Music Works: A Perspective.” If you want to learn more about generative music, start there. Tero’s work and writings on generative music have been hugely influential to me, like this tutorial on creating generative music systems in JavaScript.