A few weeks back, I was chatting with my buddy Brent Butterworth—former SoundStage! Solo editor and co-host of the first season of the SoundStage! Audiophile Podcast—about the algorithms that drive the streaming music services many of us know and love. Brent and I both subscribe to Qobuz and Spotify for our own reasons, and we both generally agree about the strengths of each service. Where our opinions diverge involves which service does a better job of recommending music that we didn’t know we were in the mood for or that we’ve never heard before, or at least not for a long time.
Brent dumped on Spotify’s algorithm and praised Qobuz’s. That’s funny because my experience is the exact opposite. Spotify is where I turn when I want to discover new tunes—or rediscover music I’ve forgotten about—based on my listening habits. That’s especially true when I’m in the car or out for a walk. With Spotify, I can very nearly set it and forget it.
Qobuz is, by contrast, my service of choice when I’m sitting in my two-channel listening room and know precisely what I’m in the mood for, or when I need a reference track for a review.
I think about that exchange between me and Brent and laugh every time some mainstream publication pitches me on writing an article about “the best music streaming service.” The best by what metric? Sound quality? Honestly, they all sound great these days—even the ones that use lossy codecs. If you’ve never taken the Spotify A/B/X test on abx.digitafeed.net, you absolutely should, if only to see what differences you perceive when comparing audio formats. If memory serves, when I last took that test a year or so ago, I was able to identify the Ogg Vorbis 320kbps stream around 85 percent of the time, but only while wearing headphones in a completely silent room, with the volume cranked to stupid levels. And I still found the differences so slight as to be negligible.
But that’s not the point.
The point is, when all of the streaming services sound great, by what metrics do we gauge which one is the best? If your main criterion is being able to press play and hear music without too much poking, prodding, and prompting, Brent and I would point you in different directions.
And the reason for that is, quite obviously, the various algorithms employed by these services—but perhaps not for the reasons you think.
The algorithm isn’t complete without you
I’ve written about music-recommendation algorithms before, specifically in defense of YouTube as a viable source of listening enjoyment. But the one thing I didn’t discuss much in that piece was the fact that these recommendation algorithms don’t work in a vacuum. They are, in a sense, constantly attempting to answer this question: “If you like X, will you also like Y?” Granted, it’s not quite that simple. In truth, the logic flows more like this: “If we know you like X, what’s the least amount of work we can do to serve you a Y that’s likely to keep you engaged and/or spending money on this service?” But you get the point. Until you provide that X input, the algorithm is just twiddling its thumbs.
But what is that X? It’s everything from songs you’ve listened to before via that service, to whether you let new songs served up to you play through to the end, to whether you hit the heart or the thumbs up or whatever positive emoji your service uses. Every time you use one of these services, you’re feeding the algorithm information it can use to decide which songs are likely to keep you subscribed for another month.
Given the different relationships we all have with music, is it any wonder that my X plugged into Spotify’s algorithm might give different results than your X plugged into the same? What’s more, given how differently I use Spotify and Qobuz, is it any wonder that one does a better job of predicting what I might like? To be frank, discovery is just about all I use Spotify for, so it stands to reason that it’s a bit more fit in this regard. Poor Qobuz rarely gets a chance to understand my tastes because I generally shut down its recommendations immediately and go off to find exactly what I want to listen to on my own.
But I don’t think that’s the whole story.
The Napoleon Dynamite problem
I think most people probably became aware of the ways in which algorithms drive our entertainment experiences in the early days of Netflix streaming. The service faced a problem early on, in that its predictive algorithms completely failed when it came to determining whether a viewer would like the indie film Napoleon Dynamite. So much so that Netflix offered a million-dollar prize to anyone who could reduce errors in its “If you like _____, we think you’ll also like Napoleon Dynamite” math by ten percent.
Now extrapolate that problem to music and imagine how difficult it must be for Spotify to say, “Hey, you dig the Polyphonic Spree, A Tribe Called Quest, Björk, the Allman Brothers, Mulatu Astatke, Weezer, Grateful Dead, and Beastie Boys, so we’re pretty sure you’ll dig this deep cut from Was Not Was.”
The thing is, Spotify has had 12 years to get to know me, 12 years of being fed Xes by me on a nigh-daily basis, 12 years of me swatting it on the nose when it peed on my rug and tried to make me listen to the Eagles.
I will say this for Qobuz: at least it’s housebroken in this regard. It has never, ever, ever played “Peaceful Easy Feeling” unprompted. That’s probably because its My Weekly Q feature only ever plays stuff I’ve asked for directly in the past, and when it does try to feed me something new, it’s so off the rails that I quickly hit the exit.
So why does Qobuz do such a consistently good job of feeding up tunes that Brent finds intriguing while it can’t do the same for me? It could be a form of the Napoleon Dynamite problem. Maybe Qobuz can just make more sense of Brent’s love of Celtic Frost and Stanley Turrentine than it can my love of KMFDM and Casiopea.
For whatever reason—and there may be many—Spotify’s algorithm and I just get along better. And that doesn’t make Qobuz or Apple Music or Amazon Music’s algorithms bad. I probably haven’t spent enough time with them to know for sure. Even I have to admit that, aside from some weird takes, Qobuz is batting a slightly higher average these days, in that out of every 12 recommendations it makes, I might not be appalled by six of them.
I think I’m going to run a little personal experiment and, for the next year or so, use Qobuz the same way I do Spotify. In other words, instead of treating it like a library of master-quality records that I dip into when I know exactly what I want to hear, maybe I’ll spend more time with its “Moods” and “Focus” playlists. Maybe I’ll smash that heart emoji with a little more consistency. Maybe, in other words, I’ll let its algorithm do its job.
Will it work? I can’t say with any certainty. But I do know this for sure: the next mainstream publication that asks me to write an article about the best streaming service is going to get a listicle with exactly one bullet point:
The best music streaming service is the one that does the best job at what you want it to do.
. . . Dennis Burger