cosine.club
electronic music similarity search engine
TODAY'S GEMS
COMMUNITY PLAYLISTS
F.A.Q
Index of /help/faq/
how does cosine.club find similar tracks?
cosine.club finds similar tracks by analyzing the actual audio, not tags, genres, or listening history. it uses the discogs-effnet model from essentia, a neural network trained to recognize 400 discogs styles from two-second slices of spectrogram. we use the embedding it produces, a vector of 1,280 dimensions that describes how a track sounds. those dimensions carry era, place and scene as well as sound, because that's what discogs styles encode. every track in the catalogue already has its vector stored. that's over two million of them. when you paste a track we run it through the same model and compare it against all of them by cosine similarity.
how is this different from spotify or last.fm recommendations?
spotify and last.fm build recommendations from metadata and what other people listened to. that works well on music plenty of people already play, and badly on everything else. a record needs tags, plays and co-listening data before it can be matched at all, so the more obscure it is, the less those systems have to work with. cosine.club reads the audio instead. a 300-pressed white label with no tags and barely any plays sits in the catalogue on the same terms as anything else.
can i use this to identify a track, like shazam?
sometimes, though not the way shazam does. shazam fingerprints a recording, which is why it survives crowd noise and phone speakers. cosine.club matches on how a track sounds overall, so a phone recording from a club won't find the studio version. it isn't really the same audio any more, just a noisy version recorded at a distance, and that's enough to move it somewhere else entirely. it works well on a clean rip. upload one to vocaroo and paste the link. if that recording is already in the catalogue it usually comes back as the closest match. if it isn't, the nearest results are often its label mates or its contemporaries, which can sometimes be enough to place the era and the scene.
what does the match score mean?
the match score is how closely a result sounds like your reference track, from 0 to 1, where 1 is most alike. it's the cosine similarity between the two tracks' vectors. a high score means they sound alike, not that either one is any good, so it's worth digging well past the top few.
is cosine.club free? do i need an account?
search and similar tracks are free and need no account. an account adds playlists, favorites, history, and an api key. there are no paid tiers. if you want to help with running costs you can become a supporter.