Music just ain’t what it used to be. At least, that’s the stereotypical lament of each receding generation of music listeners. It’s also one way to read a new study on the evolution of pop music in the past half-century.
A group of researchers undertook a quantitative analysis of nearly half a million songs to look for widespread changes in music’s character over the years. The findings, published online July 26 in Scientific Reports, show that some trends do emerge over the decades—none of them necessarily good. (Scientific American and Scientific Reports are both parts of Nature Publishing Group.)
The researchers based their analysis on the Million Song Dataset, a publicly available 280-gigabyte file that provides a sort of background sketch—name, duration, tempo, and so on—of songs from nearly 45,000 artists. Of the million songs therein, 464,411 came out between 1955 and 2010 and include data on both the sonic characteristics and the year of release.
Joan Serrà, a postdoctoral scholar at the Artificial Intelligence Research Institute of the Spanish National Research Council in Barcelona, and his colleagues examined three aspects of those songs: timbre (which “accounts for the sound color, texture, or tone quality,” according to Serrà and his colleagues); pitch (which “roughly corresponds to the harmonic content of the piece, including its chords, melody, and tonal arrangements”); and loudness (more on that below).
After peaking in the 1960s, timbral variety has been in steady decline to the present day, the researchers found. That implies a homogenization of the overall timbral palette, which could point to less diversity in instrumentation and recording techniques. Similarly, the pitch content of music has shriveled somewhat. The basic pitch vocabulary has remained unchanged—the same notes and chords that were popular in decades past are popular today—but the syntax has become more restricted. Musicians today seem to be less adventurous in moving from one chord or note to another, instead following the paths well-trod by their predecessors and contemporaries.
Finally, it comes as no surprise that music has gotten louder. A piece of music’s loudness is an intrinsic characteristic of the recording, not to be confused with the listener-controlled volume. “Basically, the audio signal, when recorded and stored, is physically bounded to be between certain values (+1 and –1 volts in original recording systems),” Serrà explained in an email. “You can record signals fluctuating between –0.2 and +0.2 or between –0.6 and +0.6 (positive and negative fluctuations are necessary to make the loudspeaker membrane move). That's the intrinsic loudness level we're talking about.”
For years audiophiles have decried the “loudness wars”—the gradual upping of recorded music’s loudness over time, in an apparent effort to grab listeners’ attention. Loudness comes at the expense of dynamic range—in very broad terms, when the whole song is loud, nothing within it stands out as being exclamatory or punchy. (This two-minute YouTube video does a great job of demonstrating how excessive loudness saps richness and depth from a recording.) Indeed, Serrà and his colleagues found that the loudness of recorded music is increasing by about one decibel every eight years.
It’s an interesting study, and it seems to support the popular anecdotal observation that pop music of yore was better, or at least more varied, than today’s top-40 stuff. (A recent study also found that song lyrics are darker and more self-focused than they used to be.) But I did wonder if there was a selection bias in play here. The Million Song Dataset, huge as it is, may not provide a representative slice of pop music, especially for old songs. Its contents are heavily weighted to modern music: the database contains only 2,650 songs released between 1955 and 1959, but nearly two orders of magnitude more—177,808 songs—released between 2005 and 2009. That’s because it draws on what’s popular now, as well as what has been digitized and made available for download. And the songs of yesteryear that people enjoy today (as oldies) may not be the same ones that people enjoyed when those songs first came out.
Let’s assume for the moment that the trends identified in the new study—especially the homogenous timbres and restricted pitch sequences—are bad. Then the rare song that bucked those trends, offering up novel melodies and sonic textures, would stand out as being good. Therefore, that song would have a better chance than its contemporaries of surviving the test of time—that is, a better chance of finding itself digitized and widely played some 50 years after its release, thereby boosting its odds of inclusion in the Million Song Dataset. Meanwhile, the blander tunes of decades past would have faded into analog obscurity. The relatively few old songs in such a database, then, would tend to be more sonically interesting than the average song of today, and any analysis comparing old songs to new would likely reflect that. So I wondered if part of what this study is telling us is that bland music can fool us, but not for long.
Serrà acknowledged in an email that a bias due to the “test of time” effect is possible but argued that its influence should be small. For instance, he noted, the long-term patterns and trends that he and his colleagues identified also hold over relatively short—and relatively recent—time periods (say, 1997 to 2007), where the “test of time” effect should be minimal. “The same happens with close and not-so-recent time periods (e.g., 1960 and 1968), where both years could partly incorporate such an effect,” he wrote. “Since the trend is consistent in short time spans where you assume the ‘test of time’ bias is minimal and, furthermore, the trend is also consistent for longer time spans, we can assume it is a general trend and, thus, that the ‘test of time’ effect is really small.”