Making love with a tambourine

Mix 37: Some thoughts on data alchemy and genre, checking in on Caribbean pop, spotting a few fun fakes, and two favorite 2024 South African artists return

Whatever you call the thing that The Who, Boston, Guns N’ Roses, Led Zeppelin and Lynyrd Skynrd share, Rihanna has it and will also blow up your house with her car.

As long promised, I finally finished “Spotify data alchemist” Glenn McDonald’s You Have Not Yet Heard Your Favorite Song, an accessible, if somewhat impressionistic, collection of essays on algorithms and data that’s worth reading even if you end up hunting and pecking. The book doesn’t go very deep into data nerdery, only gestures toward it based on the themes of its many short chapters. At some level this is fine, since when it does get nerdy, I get side-swiped by math.¹ But I would love to see a more detailed theoretical visualization (i.e. not proprietary information, which he isn’t allowed to share) of the vector embedding that underlies the social and genre mapping at Spotify.

The idea in vector embedding is that you turn lots of a given song’s attributes and metadata into numbers: a song’s popularity on the platform, its individual sonic features (e.g., beats per minute or “banjo-ness”), who is sharing it on which playlists from which countries—and, interesting to me, which songs are uniquely popular in that country, forming an account of regional listening that doesn’t just reproduce global streaming data. Then you see how all those numbers—sounds, shares, streams, etc. —correspond to other songs in a topography, with many different paths for one song to be “close” to another song. I think I got the concept, but I wanted to see snapshots of this taste landscape.

There are important points woven through the book: algorithms are much dumber than you think, even with so-called machine learning; most of a categorization tool’s effectiveness is built on human curation and decision-making; prioritizing attributes and maintaining editorial control of the categorization system is important, because if you leave it up to pure folksonomies (i.e. people tagging their own genres and information however they like), you’ll quickly get a disorganized mess. There’s a cool archival science book hiding in there.

But you also get comparatively weak treatments of more political and philosophical concepts like cross-cultural interconnectedness or the nature of genre. I got inordinately hung up on this comment on pg. 186, in which McDonald describes the process of creating a “rock radio” filter for a client:

“Whatever you call the thing that The Who, Boston, Guns N’ Roses, Led Zeppelin, and Lynyrd Skynrd share, Rihanna doesn’t have very much of it.”

This is a minor aside in the scheme of things, and it is an accurate assessment of what McDonald’s client is looking for, which is a rock song recommender that doesn’t just give you Rihanna. But it rankles.

I find some of this fine-grain slicing of genre annoying. It feels arbitrary, and at a technical level it is: lots of weird-sounding genres are translations of numerical values based on observed taste maps. The act of modeling the topography—letting the data suggest to you which songs are “close together”—ultimately constructs genre from the perspective of the machine. When you understand the mechanics of how real communities are translated into numbers, it makes sense. But the resulting proliferation of hyperspecific microgenres can be treacherous as a user: at times it reminds me of Rate Your Music, which is so overrun by garbage categorization that it’s functionally unnavigable, often making exploration seem harder than just sorting the firehose of data yourself.²

I’m also a big believer that genre’s role in music listening and discourse and life-living has to include at its core serendipity and contestation, both of which seem like potential casualties of the approach to categorization that animates a lot of McDonald’s writing, where serendipity is managed and contestation is redirected to a new and separate branch in the category architecture. There’s a telling anecdote where McDonald describes how helpful the METAL section at the music store was in sorting out what he was looking for as a young, pre-internet music consumer.³ He goes on to say that with Spotify, you can now have a section just for gothic symphonic metal.

But this potentially erodes the provocative social value of genre, which doesn’t just reflect communities and their parties, but dares everyone else into the party, too, perhaps in the very process of trying to keep everyone out (and thus conveniently sorted). You can maybe ask “what the hell is Teena Marie doing in the METAL section?” but no one’s ever going to sort her under gothic symphonic metal. And she definitely won’t get in if the program is designed for her numbers never to match.

Collapsing the distinction between genre as a battleground/playground and as terminology designed for descriptive precision seems misguided to me as an intellectual project, even if it impresses as a feat of engineering. It helps you build an incredible machine, but it also just doesn’t sit right to keep “Fire Bomb” by Rihanna out of rock radio, even in the age of the celestial jukebox. Whose utopia is this anyway?

---

Previous 2024 mixes

MIX 37: MAKING LOVE WITH A TAMBOURINE

1. Shaydee’s f. Elji: Soum Soum

2. Kryssy f. Mikado: Anlè an jet

3. Meryl, DJ Tutuss: Jet Ski

4. DJ Kawest, T-Gui, T-will: Konpa Paradise 2

5. DJ Kawest, Ayewai: Whippin

But enough of my griping, the machines are good, too. As a testament to Spotify’s underlying functionality, here’s a mini-playlist of Caribbean pop put together with the help of some of that music model magic—a combination of the right genre inputs (zouk riddim, kompa, kompa goyad, bouyon, raboday) and algorithmic assists to find new playlists. But this dive (not a very deep one) was sparked by IRL recommendations from Jonathan Bogart, who has been my go-to intermediary for tropical pop music for years.

It’s a nice use case for the genre machine. There is some inconsistent tagging, so a lot of songs don’t show up in a genre search, and meanwhile sparse human-curated regional lists can be all over the map (literally). In combination, though, you do end up with a decent cross-section of music from the region to get started—in this case, a few hundred songs from 2024. It would be hard to find this stuff without some algorithmic nudging, either from Spotify or, perhaps more effectively for this scene, from YouTube’s “people also watched” suggestion generator, which works pretty well if you’re looking for certain regional music even though it’s often terrible for general interest topics.⁴

Some things that I noticed in my wanderings this week: a playful form of dembow has made its way to other Caribbean music over the past few years, represented here by Martinique artist Meryl, whom I featured in a previous mix. Also from Martinique: wonderfully sharp minimalism in tracks from Shaydee’s and Kryssy that seems to have some of Brazilian funk’s logic of bold juxtaposition, but with a softer rhythmic backdrop that’s not a world away from my favorite minimalist diva R&B from the mid-aughts (Cassie, Christina Milian, Natasha).

There’s also more brazen sampling and interpolation than I remember from previous Bogart roundups. I didn’t even include the two tracks Jonathan shared that set me on this journey to begin with—Guadeloupean bouyon artist Daly’s “Lala,” which samples P.M. Dawn, and Franco-Haitian singer Beverly Bardo’s “Curaçao,” which quotes “Hotline Bling.” I went instead for two tracks by French-Guadeloupean producer DJ Kawest, who made me gasp when I heard what he did with the tumbi hook from “Get Ur Freak On” and made me laugh when I heard what he did “Für Elise.”

6. Comoriano, DJ MG974: Dora

Francophone rap from Comoros that came from one of the Carribean playlists and about which I didn’t find a ton of information, but sounded good leading into (and is the same key as) the next song, softening the perhaps inevitable whiplash of moving on to…

7. Ava Max: Spot a Fake

I spotted a fake and I liked it: Ava Max has emerged as the fakest and maybe best of a crop of pop stars that folks who have not yet cheekily adopted the term A-Pop sometimes refer to as the “pop girlies.” She threatens true cringe with all of those ‘80s sounds that her too-damn-tasteful peers won’t touch—real training montage shit. If she didn’t curse at the phony new girlfriend, I’d even play it for my kids.

“My Oh My” beat “Espresso” handily in the “Song of the Summer That Was Featured on Mix 14” sweepstakes in my household. My kids still don’t really know any of Sabrina Carpenter’s music, but are terrified of her after accidentally seeing the first minute of the “Taste” video at a friend’s house before panicked parents could intervene. Direct quote: “That woman kind of haunts me.”

8. Galantis: 8 Days

Alright now, fellas: what’s faker than being fake? Galantis!

9. Ana Mena, Emilia: Carita Triste

One of my favorite Argentine pop artists duets with a Spanish singer that I have finally managed to tell apart because she is the blond one in the video.

10. Tuấn Cry, Masew: Giai Điệu Việt Nam Mình

Here my extremely spotty knowledge of Vietnamese pop trips me up completely, as I feel totally unequipped to describe how this song uses deceptively minor drill ornaments to springboard from a traditional melody into the present, let alone hazard a guess as to whether it’s a big deal that it’s doing so. I’m also curious if the Google translation of the lyric “compliment someone who is skillful in making love with a tambourine” is anywhere in the ballpark of accurate—different attempts have it as “patting love” or “beating love with the rice drum.” But I’m going with it.

11. El Khat: La waLa

Berlin-based Yemeni band with some solid percussive Middle Eastern psych. Whole album is good.

12. Dj Helviofox: Cala a Boca

The latest batida “fox” does not disappoint, though I can only use adjectives like “loping” and “insistent” so many times before I give up even trying to describe this stuff.

13. DJ NBZ, Yuri Redicopa: Sonho Ritmando 2.0

Baile funk pick of the week has some fun with what sounds like a bailalaika. (It may be some obvious sample I’m missing—sounds like something from The Godfather.) Recently reorganized my holdover folders of regions in ferment with too many songs to include on my mixes (in decreasing playlist size order, Brazil/funk, South Africa/amapiano, and DR/dembow) so if you need more funk options, there are about 300 more. And those are just the ones I find on Spotify, which I’d guess is far less than half of what’s worth listening to.

14. DJ Warlord: Begging for an Encore

Finding a PC Music track in a blind taste test really underscores for me of how much of the group’s success was branding; I don’t see this as categorically different from any of the other fussy, hyperactive dance-ish music I stumble across, but it’s a good ‘un.

15. Cities Aviv: Friends

Memphis producer creates (curates?) the sort of lost 80s quiet storm also-ran you’d find on a moldy old cassette at a garage sale, but I couldn’t identify the sample assuming it is one (a lyrics search only pulls up this song). If it’s not one, I’m even more impressed.

16. Xamd: あなただけが

A belated brat summer entry—maybe in Japan it’s brat autumn.

17. Tota, Daoko: Addiction (Esme Mori Remix)

Assumed the elements I liked, like the little Wii-core drops in the chorus, are from the remix, but the original is pretty close, had stronger bones than I expected. Reminds me of finally hearing and liking the original “Something Good Can Work” by Two Door Cinema Club after assuming nothing could have been close to the Twelves remix.

18. Zee Nxumalo, TBO f. PYY Logdrum King, DJ Tearz, Dr Thulz: Ngisakuthanda

19. Atmos Blaq: Skhaftin Saka Fonque

Sweet, I have 14 minutes to spare for an amapiano double-header. First, some trad-ish amapiano from my favorite South African singer of the year, Zee Nxumalo. Then more 3-step from Atmos Blaq, a few months after already putting out a contender for South African song of the year, “Mfana Wase Dobsi.” Zee Nxumalo has another contender with “Thula Mabota.”

20. Jungle by Night f. Pitou: Dive

Ending with an cool-down sequence, starting with some moody Dutch electro whose Kraftwerkian impulses are mitigated by lovely vocals from featured singer Pitou.

21. Stuzzi: Solåmåne

Next: light and vaguely tropical electro-pop from a Swedish producer I am surprised to have featured once before (I recognized the cover art on his artist page), and about whom I previously wrote “throws as many signifiers as he can at the wall and a bunch happen to stick.” Sure.

22. Sans Froid: Planket

The home stretch! Scuzzy Bristol indie with sweet vocals, piano plonks, and maybe a dash of Tori Amos.

23. Phi-Psonics: Arrival

And finally, a diaphanous jazz closer. Flute!

***

That’s it! Until next time, don’t put all your songs in one basket, but also don’t put them in a thousand different baskets.

—Dave Moore (the other one)

Title translated from Tuấn Cry & Masew’s “Giai Điệu Việt Nam Mình” (“Khen ai này khéo vỗ tình bằng trống cơm”)

1. “Vector embeddings…let you measure the distance between things by taking the square root of the sum of the squares of the differences between corresponding numbers in the lists for the two things” (pg. 157) did not give me the sort of wonky detail I was looking for. Though it was amusing to re-read it several times in the voice of the Scarecrow at the end of Wizard of Oz to try to parse it.

2. The difference is that RYM’s categories are user-generated, a folksonomy built by annoying music pedants. But some of Spotify’s editorial genres to represent “song proximity” can seem this way, too. I usually find RYM to be a minefield, a Borges parody of a music database. But sometimes you do find an annoying little box that leads you to something massive. This is a good footnote to emphasize that a lot of my gripes here are pet peevish.

3. I suspect that McDonald’s being a metal guy explains some of his penchant for what seems to me like occasionally counter-productive genre hair-splitting.

4. Spotify has been a poor aggregator for Caribbean pop, largely owing to the slow roll-out of the platform to many countries in the region (many finally got it 2021; Cuba is still a no-go). But lots of the songs are available if you know where to look for them. This is part of what makes the mapping so important—you can’t really trust curated playlists to get you started, but once you’ve cracked some of the key artists and genre tag conventions you’ll start getting better leads.