Tuesday, June 28, 2011

Loudness War Research

There's an interesting article up on Digital Music News about the Loudness War in popular music (i.e. the tendency for released music to have less and less dynamic range over the years). While this is a topic that has been written about quite a bit over the past several years (including a lot of bellyaching from yours truly), most of the assertions about loudness and popular music are merely anecdotal in nature with little hard data to back them up.

Two graduate students at Rutgers University, Shaun Ellis and Tom Engelhardt, are looking to change that, having recently released a trove of data related to chart-topping songs dating back to 1960:

They scanned an exhaustive range of attributes related to top-ranked songs, including tempo, time signature, key, and chord progressions.  The Echo Nest offered quite a bit of data, algorithms, and API-related assistance.   
Unsurprisingly, lots of hit songs share common attributes, especially elements like verse-chorus-verse and 4/4 time signatures.  But since the 1960s, there's been a clear march towards louder, fuller, more blaring music. 

Ellis & Engelhardt chart the rising volume of popular music. (From Digital Music News)

The chart above shows Ellis and Engelhardt's data relating to the average dB of popular songs from 1968 to 2010, and the relatively linear increase in volume is clear.

The chart is interesting to me because if the data is accurate, it does not support many of the conventional assumptions about the loudness war (including my own). The main culprits in most explanations as to why music has gotten louder over the years are usually related to technology, and especially the advent of digital recording, mastering and playback. However, the data here doesn't fully support such explanations. If anything, you can see a slight (albeit temporary) dip in volume levels during the early years of the CD era (the years 1990 to 1998 are the only sustained period where the actual data falls below the averaging curve). And the biggest jump occurs between 1998 and 2001 just before the iPod was introduced. If anything there's been a slight dip in volume levels since the introduction of the iPod in 2001 (which was the loudest year for popular music according to Ellis and Engelhardt's data).  Looking at this chart makes me think we ought to be cautious in accepting any explanation that blames a particular technology (be it CD, CD changer, iPod, etc.) because the rise is fairly consistent across a long period of time that saw various technologies come and go.

It's also interesting to note what looks like a sharp drop in volume between 2009 and 2010. This might represent a reaction to the numerous complaints about the loudness war from consumers, music critics and bloggers, or it might be a momentary blip. Given the consistent rise in volume since 1968, despite year to year variation, at the moment I would consider it nothing more than a bit of statistical noise. I'm certainly not ready to declare the loudness war "over" based on what I see here.

Now with that said, there are a number of questions about Ellis and Engelhardt's methodology that I would want to know the answers to before coming to any serious conclusions based on their research. First and foremost I would want to know how they measured the average dB of the songs they selected, as well as the methodology for choosing which songs to measure in the first place. At the moment it does not appear as though they have released this data. It's fairly easy to get data like this from a CD release, but it's much harder to get it from an LP or 45 RPM single. Further, the same songs were often more compressed (louder) on 45 RPM single releases than they were on LP versions. Which versions did they choose to measure, and was this done in an organized and consistent way?

Another thing to consider is the data represented in this graph only apparently relates to hit songs, not to reissues of older music, which is one of the things about the loudness war that bugs me most. It's one thing for new hit music to be recorded, mixed and mastered to have little dynamic range, but it's another thing when older music gets the dynamic range sucked out of it to keep pace with current trends. The data also presumably only relates to charting songs, not full albums or less popular music.

Nevertheless, whatever its limitations, Ellis and Engelhardt's data is interesting, and provides some welcome quantification to a phenomenon that has been greatly discussed, but is probably still not well understood.

No comments: