The Dante’s Purgatorio concept album I finished a few months ago has an introduction track on it that functions as a slow-building cinematic ambient drone piece before the “real” songs start. It’s been interesting to see these it show up in relatively high numbers on the stats that Apple (who owns Shazam) sends out each week. Here is what last week’s report looked like:
At first glance, it’s “Wow! 15 people were listening to my songs last week!” But then I look at the breakdown. Nearly every time it looks like this:
Those tracks, the Prologue, Epilogue, and “Night in that Place” are the three ambient tracks on my album. They always constitute the bulk or entirety of the Shazam stats. Was some guy in Warsaw really listening to it? I’m certain they were not. During that same time I had virtually zero streams on of any of those tracks on any service. Nobody touched any of the tracks on Spotify, Amazon, iTunes, Tidal, Bandcamp, YouTube, etc. It seems clear to me that Shazam is mis-identifying the track.
I can verify this from the other direction as well. There is a coffee shop in town I frequent that often plays ambient and “chillout” music in the background. I think they might be playing some Sirius XM radio playlist. Who knows, but I will often try to Shazam the tracks from my phone for fun. What do I find? Well if the song has vocals or a clear melody it usually picks it up right away. If it’s ambient though… well, it will think about it a while and then give me some obviously wrong answer. So I try again. A different wrong answer. Sometimes I’ll Shazam it eight times in a row and get eight different answers! It has no clue.
I’m pretty sure that’s what’s happening with my ambient tracks as well. Whatever Shazam’s sonic analysis algorithm is looking for, a pile of droning synths just breaks it completely. It can’t make heads or tails out of what it’s hearing, so it just throws out garbage matches to the listener in desperation.
I’m sure that improving this is really challenging and some developer at Shazam has probably been tearing his hair out trying to figure out how to improve the matching for this sort of thing for years. I’d love to talk to that guy actually. It would be really interesting to hear about how they’ve tired to improve it and what hasn’t worked (yet). Someone might use this as an opportunity to dunk on ambient music as being trash, but I don’t think so. It’s qualities are just difficult for computers to discern.
In the meantime, if you produce ambient or cinematic music, you might want to not get too excited about your Shazam stats, alas.
(This post was prompted by seeing the weirdness in my own Shazam stats, and by reading a recent newsletter where Robin Sloan was excited about the Shazam stats for his ambient project The Cotton Modules. Sorry to be a downer Robin! Love your work.)