When Shazam doesnt bring the magic that you need
January 31, 2019 3:48 PM Subscribe
Looking for music recognition software or an app that will listen to a song and suggest songs that are similar to it, not actually identify the song that is playing itself. Turns out this may not be a thing?
I'm aware of computer programs and apps that can offer suggestions for similar music based on an artist or song, but those only work when the song is well known enough to be on a big 'ole list somewhere. I know about things like Shazam etc. that can listen to a song and tell you what it is after playing it for a few seconds. I'm looking for something in between, where I can play a song that is too obscure (local band etc.) for a platform like Pandora or Spotify, that will listen to the music and spit out songs or artists that are similar. It's prefered that the results are of well-known bands, so sites like SoundCloud aren't real helpful here. I've googled and come up short. It may be my keywords, or maybe this just doesn't exist? Does this exist? This isn't for any nefarious purpose or 'gotcha' games if that matters.
Thanks!
I'm aware of computer programs and apps that can offer suggestions for similar music based on an artist or song, but those only work when the song is well known enough to be on a big 'ole list somewhere. I know about things like Shazam etc. that can listen to a song and tell you what it is after playing it for a few seconds. I'm looking for something in between, where I can play a song that is too obscure (local band etc.) for a platform like Pandora or Spotify, that will listen to the music and spit out songs or artists that are similar. It's prefered that the results are of well-known bands, so sites like SoundCloud aren't real helpful here. I've googled and come up short. It may be my keywords, or maybe this just doesn't exist? Does this exist? This isn't for any nefarious purpose or 'gotcha' games if that matters.
Thanks!
Don't think it's really a thing. Analysing a song to determine a style and genre without identifying the specific song sounds like it's way beyond anything that currently exists.
posted by pipeski at 4:31 PM on January 31, 2019 [1 favorite]
posted by pipeski at 4:31 PM on January 31, 2019 [1 favorite]
Spotify effectively does that for internal purposes. As early as 2014 they had neural network approaches that actually could predict which songs sound similar, or are of similar genres, just from the sound waves. But they aren't releasing it to the public.
If it doesn't exist from a big streaming media company then it probably doesn't exist, though, because nobody else has legal access to a big database of every song to train on.
posted by vogon_poet at 4:43 PM on January 31, 2019
If it doesn't exist from a big streaming media company then it probably doesn't exist, though, because nobody else has legal access to a big database of every song to train on.
posted by vogon_poet at 4:43 PM on January 31, 2019
Pandora can do the recommend a song based on another song. So, use Shazam, and then type that into Pandora.
posted by MythMaker at 4:59 PM on January 31, 2019 [1 favorite]
posted by MythMaker at 4:59 PM on January 31, 2019 [1 favorite]
Spotify does provide results of their own internal song analysis (and pretty dang solid song suggestions based upon the analyzed parameters) via their API - or they did last time I checked - but yeah they would need to accept streaming user input (with all the ambient noise that would entail), and run that same analysis in near-real-time to make it useful for what you’re asking. I wouldn’t hold my breath.
posted by STFUDonnie at 5:03 PM on January 31, 2019
posted by STFUDonnie at 5:03 PM on January 31, 2019
Pandora's whole thing is the Music Genome Project, but if it doesn't have a song already in it, it's not going to do you any good because they're all tagged manually.
posted by BungaDunga at 5:10 PM on January 31, 2019
posted by BungaDunga at 5:10 PM on January 31, 2019
No, it is a thing, I went to a presentation demoing some research that did something like this, about 12 years ago. I think they were following, as the most promising approach, kind of indexing music into a vector space of tempo and patterns of loudness and softness. I'm sorry I don't recall what the software was called or who the researchers were.
It was a presentation at JavaOne, in about 2006.
posted by thelonius at 6:01 PM on January 31, 2019
It was a presentation at JavaOne, in about 2006.
posted by thelonius at 6:01 PM on January 31, 2019
Best answer: Well - you could try Google Music. This has free and paid versions and my experience is with the former. It has a scanning app that will look over your music collection - and once this has completed, there is a “Radio” option that you can select on any track you possess - which will then produce a playlist of songs that go with that first song (the non- free version will give you a playlist of any song in Google’s collection as well as yours).
What makes this different from other services, such as iTunes’ “Genius mix”, is that you never get an “insufficient data” message on a track. Google will always have a valient try at giving you a list of tracks that it thinks are similar - even if the seed track you have given it is something super obscure that is newly released by a group of your friends. That list will be as long as you care to make it - your seed track will be repeated occasionally but each time the algorithm then go off on a new tangent - so other tracks are very unlikely to repeat. To achieve this, I believe they are using a range of matching databases including acoustic ones.
This may not be 100% what you requested- but it is easy to try out.
posted by rongorongo at 12:20 AM on February 1, 2019 [2 favorites]
What makes this different from other services, such as iTunes’ “Genius mix”, is that you never get an “insufficient data” message on a track. Google will always have a valient try at giving you a list of tracks that it thinks are similar - even if the seed track you have given it is something super obscure that is newly released by a group of your friends. That list will be as long as you care to make it - your seed track will be repeated occasionally but each time the algorithm then go off on a new tangent - so other tracks are very unlikely to repeat. To achieve this, I believe they are using a range of matching databases including acoustic ones.
This may not be 100% what you requested- but it is easy to try out.
posted by rongorongo at 12:20 AM on February 1, 2019 [2 favorites]
Came here to suggest what rongorongo said, so I will second using Google Music to identify the song, then ask it for related.
posted by token-ring at 11:21 AM on February 1, 2019
posted by token-ring at 11:21 AM on February 1, 2019
Last.fm
posted by oceanjesse at 6:39 PM on February 1, 2019
posted by oceanjesse at 6:39 PM on February 1, 2019
This thread is closed to new comments.
posted by greta simone at 4:24 PM on January 31, 2019 [4 favorites]