A while ago I shared an exceptionally useful tool called Batchcamp that I’ve used to bulk download my Bandcamp collections.

Batchcamp has been such a useful browser extension that I wanted to check for some other useful ideas the creator, hurfyd, might have developed. I was not disappointed. hurfyd has been quite busy developing new bits of software under the heading deejay.tools

The list of new applications is pretty impressive and include a Bandcamp playlist creator and a Bandcamp tempo adjuster that also works on Discogs.com (very handy for the DJs among us)

The most exciting project for me, though, is Cosine.club

Cosine.club ingests a track of your choosing from Bandcamp, SoundCloud or YouTube and then spits back an extensive list of similar tracks in a streamable YouTube playlist.

Using AI as a similarity search tool for music isn’t quite new anymore. There are plenty of playlist generators and methods from recommending similar content built-in to a variety of streaming apps. The difference I see with Cosine.club is that the search for similar tracks goes a bit deeper than I typically experience.

For example, if I use Cosine.club to search a Front Line Assembly track, then what I expect from most services is a list of recommendations based on similar artists, but what I get is a list of tracks based off the particular FLA track that I quieried. So there’s no KMFDM or Cyberaktif, or even Skinny Puppy.

Instead I get a massive list of tracks by artists I’ve never listened to, and plenty that I’ve never even heard of. These aren’t second-tier selections either.

Check out this playlist I created from an early FLA track. It’s alllll goood. Truly inspiring and has hurt my bank account as I rush to Bandcamp to further investigate particular favorites.

The trick to these exceptional results, as the Cosine.club site states, is that cosine.club repurposes a machine learning model originally designed to classify music genres, generating vector embeddings for over a million electronic music tracks. using vectors for a source track, it calculates a cosine similarity metric between the source and all other tracks in the dataset to find the closest matches

I can’t very clearly articulate that in simplified terms. All I can really tell you is that Cosine.club puts shame to the usual recommendations provided by most stream services. I highly recommend that you try it out, and when you inevitably hear something new that speaks to you see if you can find it on Bandcamp and support the artist by making a purchase.