Every Noise at Once

The terms "sameness" and "average" are doing some heavy lifting here. In the grand internet tradition I will memetically unpackage "sameness" and "average" into four related components: (1) things that are BIG (2) things that are HOT (3) things that are COOL and (4) things that are GOOD.

(1) things that are BIG — these are side effects of technological advancement, globalization, the Internet, and frictionless social media platforms. More people than ever before can book a TAP Air flight to Barcelona, stay in a budget AirBnB, and check Google Maps for the same "best cafe near me" that everyone else is going to. Access to the world and to information is more open than ever before, so more people have both the ability to congregate around the top N things (physically or virtually) and the means to do so.

(2) things that are HOT — trends have always been around and always will be. Instagram face, the Kardashian look, hoverboards, heroin chic, microservices, Livestrong bracelets, putting "F*ck" in the name of your self-help book, flat icons, frozen yogurt, Dutch tulip mania — you can localize each one of these into a time period easily enough. Popular because they are new and a momentary cultural touchstone and different enough from the n-1th state of culture.

(3) things that are COOL — what you might call "signaling theory effects". "Ads Don't Work That Way" [1] explains this in the advertising context better than I can, so go read that. Differs from (2) in that the cultural…

I sometimes wonder if we're looking in the wrong place with music recommendation systems. I've tried both Apple Music and Spotify; it's rare that I hear a song come through on the linear recommendation stream and think to myself "oh my gosh! that's exactly what I wanted!" For me, discovering new music is a branching experience, where I'm constantly listening to little bits of different things, figuring out what I like, and then looking online on forums and blogs to see what's similar to that. It's surprising that the company that owns YouTube, a platform driven by user choice and 'rabbit-hole discovery', would be looking for a new way to feed users linear song recommendations. I would much rather be able to see several 'similar songs' while listening to something, similar to YouTube's recommendation tab. Alas, no streaming service seems to have implemented this (not even YouTube music, afaik).

https://cosine.club/ is the closest I've seen to the ideal branching system. My understanding is that it uses vector embeddings to search for songs that are similar in sound, and it works shockingly well for that purpose. However, it has a limited song database. Also see https://everynoise.com/, which is no longer updated. These use vector embeddings in similar ways, but the exploration experience is controlled by the user, not by a list-generating ranking model. I definitely think that AI-tech is the future of music recommendation, but I would prefer to see more research by large…

I mostly use playlists generated by this service https://everynoise.com/ (you can get to them with desktop client by pasting playlist links from the site to the spotify client search box.)

From the site:
"Every Noise at Once is an ongoing attempt at an algorithmically-generated, readability-adjusted scatter-plot of the musical genre-space, based on data tracked and analyzed for 5,646 genre-shaped distinctions by Spotify as of 2021-10-16. The calibration is fuzzy, but in general down is more organic, up is more mechanical and electric; left is denser and more atmospheric, right is spikier and bouncier."

I personally think that an algorithm trying to learn my tastes and recommend music to me is the wrong approach... especially as the algorithm does not actually seem to care anything about the actual waveform content.

I feel I get get this strange "personal echo chamber effect" where I at least think I notice the effect of my past listening habits or the outsized weight of some stuff that was on powerplay some time ago. It gets hard to find new stuff or to learn to like new stuff.

This varies though, sometimes the recommendations are great. Perhaps they work best for me if I don't constantly follow or listen to them, but instead listen to playlists curated by humans or everynoise.com.

Spotify has not viewed itself as a music company for longer than that. It's a platform for audio. And, while there are still music first people at the company, they are not in the power positions that they used to be.

The transition didn't start when they laid off Glenn MacDonald, but that sort of cemented it. They had already gutted curation before that and by this time you were far more likely to find people talking about AI in the halls than music. If you've never heard of Glenn, check out his book: "You Have Not Heard Your Favorite Song: How Streaming Changes Music." Or his old online projects at https://everynoise.com/.

We're not likely to see one. DALL-E was trained by analyzing pixels along with the captions for images. It is the caption information which is used in understanding your prompts.

But music doesn't have a form that ties the audio portion so cleanly to a textual description. And when it does, those labels tend to be overly simplified and not really helpful. Music is, in fact, hard to describe.

One project that might interest you is Every Noise at Once[0], which does an amazing job of grouping known artists and songs by their sonic similarly, which results in similar style and listener appeal.

[0] https://everynoise.com/

Ya'll should checkout https://everynoise.com/. Similar in spirit.

Just read the update:

> 2024-01-05 status update: With my 2023-12-04 layoff from Spotify I lost the internal data-access required for ongoing updates to many parts of this site. Most of this, as a result, is now a static snapshot of what, for now, will be the final state from the site's 10-year history and evolution..

what a shame. I didn't realize the author worked for Spotify. Guess it makes sense. Spotify should've acquired it from the author or made a deal with him to keep it live since all the links lead to Spotify anyway.