Rhodes is a vocal dead ringer for Paul McCartney.
SPOTIFY MUSIC DISCOVERY FULL
Zambian rock music from the ’70s is full of the strong riffs and fuzz-pedal effects you’ll hear in the Rolling Stones or the Who. They were off the beaten path but not imposingly so, often just a short jog away from music that was much more famous. Velardo identified the goal of a music-recommendation algorithm as “striking a balance between familiarity and novelty,” and it was easy to see how a robot could conceive of the songs I’d been hearing as hitting the sweet spot for people who were into vaguely dad-oriented, sorta-basic rock. My sense was that “Discover Weekly” is creating an unofficial canon of cool music and offering it to me, music supervisors, and cousins the world over. There are millions of songs that could be served up to Spotify listeners at any time. Ignoring this quite reasonable hypothesis, I pressed forward with my pet theory. So I called up music AI consultant Valerio Velardo, who suggested whatever was happening was largely taking place inside my head: “It’s selective attention.” It was reluctant to share any data that would have shed some light on why it seemed these songs were following me around - except to emphasize that the app definitely was not listening in on users’ phones. The company also uses a technology that trawls internet text and users’ playlists for descriptions of songs’ content. Mounia Lalmas-Roelleke, the streamer’s senior director of research, told me the algorithm takes into account things like the time of day, what historical era you seem to prefer, and how open a listener is to styles they haven’t heard before. But the gist is what’s known as collaborative filtering, the bedrock of most recommendation engines: If you and I both like X and I also like Y, the algorithm will serve you Y as well. The company wasn’t going to reveal each of the “thousands” of factors that go into its algorithm, which would have ruined its business and also taken all day. What was going on? In search of more knowledge, I asked Spotify to give me a general explanation of the way “Discover Weekly” works. Spotify had killed the snobby record-store employee and taken his place. Both “Discover Weekly” and my local wine shop played “Dry the Rain” by ’90s Scottish alt-rock group the Beta Band, famously name-dropped by John Cusack’s character in High Fidelity. Last fall, I was staying with my cousin in Toronto when out of nowhere she put on “Khala My Friend.” In September 2018, the streamer introduced me to Art Garfunkel’s cover of “Waters of March” three years later, the Norwegian film The Worst Person in the World used the track over its closing credits. Recently, I have felt a sense of déjà vu from the realization that my secret Spotify songs are someone else’s as well. For the most part, these are semi-obscure tracks that, because I am mildly uncool, I am hearing for the first time - foreign music that sounds western, vault tracks from artists who were little known in their own times, a depressingly large number by singers who died under tragic circumstances. Whenever I hear one I like, I save it to a playlist since I started back in May 2017, I’ve collected more than 370. 1 hit came on the Swiss charts “Good Time,” by Donnie and Joe Emerson, two Washington State teenagers whose 1979 home-recorded album was essentially unheard for decades and, at least twice, “Somebody Made for Me,” by the singer-songwriter Emitt Rhodes, once hailed as the “one-man Beatles.” It was a motley collection of tunes, but I knew them all by heart - because at one point or another Spotify had served them up to me on my “Discover Weekly” playlist, a set of personalized music recommendations updated every Monday. Over the course of the night, the bar played “Khala My Friend,” by the ’70s Zambian rock band Amanaz “Like a Chicken,” by WITCH, a more popular Zambian band from the same era “Red Lady,” a B-side by psychedelic rocker Phil Cordell, whose only No. A few weeks ago, I was at a Brooklyn cocktail bar called the Great Georgiana when I heard something strange.
0 kommentar(er)
