(Documenting my research and suggesting to disable AcoustIDs)
Stole is a song I obtained in 2005 and didn’t care at the time to take enough notes about. I only have the title and artist name as given. The recordings linked to the AcoustID indicate that the audio either represents the album version or the Pop edit. It should only be the one or the other, which means some of the recordings linked should be disabled.
Without more metadata, Picard of course has no way of determining which one it is. So, I played my version against the album version on Simply Deep, available on Spotify, and the version on album R&B - 100 Hits, also available on Spotify.
My version matches the latter, which means it is the Pop edit, which is up to 14 seconds shorter.
Up until 2:42, the Pop edit is identical to the album version. It has been shortened at that point. The ending has also been shortened.
Should I just go ahead and disable the non-“Pop edit” recordings linked to that Acoustid?
Actually, it looks like none of the recordings should be disabled.
Recording a523… has the same length as the Pop edit but the track is probably simply not titled “Pop edit” on the actual track lists. Maybe we should add a “Pop Edit” in the disambiguation comment? But I couldn’t find some of the albums on Spotify to make sure (more research required).
Those long ids are likely from older recording merges. It is especially common for the same recording to appear on many Various Artist compilations. These may initially be added as separate recordings in MB. Then an editor notices these duplicates and merges them.
In those cases all the old MBIDs now redirect to the chosen Recording’s MBID. Notice how when you click on 426d0 it opens up 937da. AcoustID can be a little slower with their cleanup so you have a few months where you still see the long numbers.
Sorry. Misunderstood your note. If anything is added or changed then they are certainly separate recordings. Sometimes if a change is subtle, or later in the track, then AcoustID does not pick it up.
But also you get cases where recordings have incorrectly been linked. When I clicked on that Stole (Dreambrotha mix) I could see two singles and two VAs. Those two VAs didn’t have that mix name and could be incorrectly linked. These could be where the other AcoustIDs are coming from.
When clicking on the fingerprints linked to that recording it looks like a clear case of some confusions going on. I am no expert on this artists work, but I agree with you that some tangled references are happening.
It can be hard to really untangle something like this without some example tracks to pull AcoustIDs from.
Just a reminder that AcoustID only fingerprints a maximum of the first 120 seconds (2 minutes) of audio of any recording.
If you take a four minute long song, and chop the last minute off the end of it to create a “new” three minute long pop mix. AcoustID will fingerprint both of these as exactly the same song/audio, because the first two minutes of audio are exactly the same.
No, AcoustID also considers the length of the audio and if the difference is more than X seconds (20? 30? – I believe the 14 seconds here are lower than that threshold anyway) it will make a new AcoustID even if those first 120 seconds are identical.
Case in point, I just came across Sky, by Sonique: When comparing AcoustID f3ac7 with 0383b, you can see that the AcoustID system groups the fingerprints from one extreme (4:27) to 4:20, and from the other extreme (4:16) to 4:23. The difference in both cases is 7 seconds.
However, I think that both AcoustIDs very likely represent the same audio.