FYI.

This story is over 5 years old.

Tech

A Track-By-Track Review of Computoser's Algorithmic Pop Songs

Two stars.
​Image: YouTube

​Computoser, as one might imagine, is a system designed to write and perform new songs on the fly, using an algorithm that expands on a selection of IRL pop and classical music samples according to some research-driven parameters. For example, a song should have both repetition and surprise. Dissonance can be good, but not always. Good songs have structure and rhythm, but, again, not too much.

The algorithm works by first calculating a "main part," which is a triad of pitch, note length, and variation (motifs should vary through time). This main part acts as the seed for everything else, which is generated sequentially. In a sense, it's computer improv.

Advertisement

I spent probably too much time listening to Computoser's output this afternoon, so I offer a review of a Computoser EP, written and recorded mostly within the past hour

"Warmth with Gloom"

These are the best Computoser tracks. Clattering, "dissonant" drums and snippets of mathy, evolving-in-time riffs. That said, the artist's obsession with the oboe or digi-oboe remains highly annoying, and for algorithmic composition, focusing on orchestra instruments begins to seem a bit "doth protest too much." Or at least we could expect better synthesized orchestral sounds, which exist and are improving at the same rate at which computer processor speeds are improving. I suppose there are additional limits imposed by having the composition done in real-time via internet.

"Leave with Overwhelming Light"

There's just nothing to this one. It's bad hold music, robo-Kenny G.

Computoser is based on probabilities gleaned from actual human-composed music. As an accompanying paper (​posted to arXiv) explains, "500 popular songs details were provided by HookTheory.com. Additionally, 50 freely available MIDI files of popular classical and modern composers were analyzed." Such raw materials, used to derive the probabilities of different music events and characteristics occuring, might help explain duds like this one, which, to its credit, has a pretty good title.

"Pieces with a Rose"

OK, now we're back. Another percussion-based track, featuring a synthesizer-as-synthesizer and a reasonably inoffensive guitar-ish lead instrument. I can handle this. Like a lot of these songs, it makes me think of slowed-down Musak Deerhoof.

Advertisement

"The Piece with a Lie"

Dislike, dislike. Perhaps part of the problem is the mixture of classical motifs with pop. That might be fine in some scenarios, but the classical contribution here (and most everywhere else within the Computoser universe) is clearly Western pop-classical canon, the stuff of Hollywood movie scores and middle school band classes.

That's disappointing because there's really a whole universe of classical music that is interesting, dynamic, and "new." I don't sense a whole lot of Penderecki or Gorecki (or even Glass) in here.

Otherwise, there seems to be an inverse correlation between title and song quality developing.

"We Go to Warmth"

Computoser does unexpected things, which is part of its algorithm. No percussion here, just that piano that I think is supposed to sound like a guitar smooshed together in interestingly dissonant ways. It could be free improv.

Here are two of the system's overarching guidelines:

Variations – simple repetition tends to be 'boring; therefore a set of variation techniques need to be employed in a piece.

Dissonance and syncopation – although music strives to be consonant and rhythmic, unexpected dissonances and syncopation can make it more interesting, therefore this element is also encoded in the algorithm.

"Glamour Kisses the Pianos"

Another great title, another unlistenable not-song. Musak suitable for neglected, high-vacancy suburban shopping malls.

The developers of Computoser tested out their system IRL, polling listeners as to whether or not they liked a given track. They found a 58 percent approval rating, which is a bit better than my own experience, but not by much.

I'll stop there because I don't see this going anywhere all that new, even though every new track will actually be brand-new right then and there. Which is the contradiction.

It's like trying to make random numbers; the system of creation, the parameters, is always lurking not too far underneath waiting to be discovered. This is the threshold of boringness.