"It's Like Slipknot Meets Bach": Why Music Is So Hard to Describe

While it's not really clear who first said that "writing about music is like dancing about architecture," the sentiment is obvious: when it comes to music, language comes up short.

Not content to be defeated by a maxim, a team of European researchers is trying to devise a framework for talking about music accurately and succinctly. If talking about music really is like dancing about architecture, then the group at KTH The Royal Institute of Technology, Stockholm, is teaching the first steps. Their work was just published in the Journal of the Acoustical Society of America.

Of course, we already have a whole lexicon for talking about music. You can describe its melody, you can talk about its tempo, you can even get into chord structure, put it into a genre, go on about instrumentation, position a piece within musical history, and so on.

With the exception of genre and musical history, most of these terms approach music from the point of view of music theory, which is certainly useful for composing and communicating ideas among music makers, but can leave the layperson out in the cold. The music theory lexicon just isn't adequate for describing the human experience of hearing music, which is probably why much of music journalism has moved away from it altogether.

To better describe what we mere humans perceive, the researchers selected nine perceptual features that describe a music's overall properties, in order to give a "higher semantic description" of the music. Two sets of about 20 people individually listened to both polyphonic MIDI versions of familiar songs, and clips of film music and then rated all features, and—in the first experiment—also the emotion descriptions.

They found that, depending on which qualities they were rating, there was a lot of inter-rater agreement. "Speed" rated from "slow to fast" proved easy to rate, as did "rhythmic complexity" rated from "simple to complex" and "rhythmic clarity," rated from "flowing to firm." The was less consensus when rating "brightness" but this study is still an early step towards a clearer way of talking about music.

I reached out to Anders Friberg, the study's lead author, to better understand what a higher semantic description is and why we'd want such a thing.

"By higher level semantic description we mean something that can be characterized in a few words , something like the final judgement of listening," he replied via email. "It could be emotions, mood, if you like it or not, genre etc. Thus it is not related to basic parameters like tempo or melody."

Friberg explained that given that people can perceive the emotional timbre of a song within a fraction of a second, "the music theoretic point-of-view of detailed analysis" isn't always the most direct way of understanding how music is perceived and therefore how could or should it be categorized?

"We have the double agenda of both trying to contribute to basic understanding of music perception and contribute to the applications within music information retrieval," Friberg said.

And one of those applications within music information retrieval could even be better internet radio stations. Spotify, for example, already has mood-based playlists. Turns out Friberg and his team worked pretty close to Spotify's office, and the music streaming service was founded by a few Royal Institute of Technology, Stockholm alum, and also this idea had already occurred to Friberg.

"I definitely think that it can be useful in the future for example for categorizing music and this is one of the aims with this project," he said. "However, at this time it is still too early for that. We first need to make models for the most interesting perceptual features. We are currently working on that and our PhD student Anders Elowsson made a model of speed that could predict the perceptual judgement with more than 90 percent accuracy."

Because, of course, we humans don't need help judging the emotional aspect of music: "In fact, this is more or less the essence of music listening in general," Friberg said.

But our computers do, at least for now. We could maybe use some help articulating what it is about, say, "Under Pressure" that lifts the human spirit. We could use some pointers for dancing about architecture.