There are many recordings which seem to be the same except for a small length difference, eg.
How do we know when to merge these without any additional recording info?
It’s a case by case decision, but in this case I would definitely merge. Look for fingerprints. The fingerprint of the 4:01 recording is already part of the other one. You can try to compare the AcoustIDs from these tracks, but with this it’s quite obvious. But be aware that there are sometimes different versions which may look similar (e.g. explicit and clean).
5 Likes
In these cases there should be a disamb field entry, but I guess it’s just an illusion to think that this is always done correctly 
3 Likes
It is easier to merge Various Artist collections like that as it is unlikely a band will record something different for a compilation. Though I have seen cases of something like a charity album getting special recordings.
3 Likes
A one second difference may also just be an artifact of different rounding methods (some tracklists round their track durations towards 0, while others will display 4 minutes and 600 milliseconds as 4:01).
6 Likes
Be very careful with AcoustID fingerprints. I have multiple examples of recordings of different lengths being grouped together, despite significant differences in the first 120 seconds:
https://acoustid.org/track/aef83d93-e4da-43c2-b0fd-c5277162e075
https://acoustid.org/track/af660683-4ca9-4f87-83bb-7c023e9419f5
Here’s another with a significant length difference, but the audio differences lie beyond the first 120 seconds:
https://acoustid.org/track/a0fb4bc2-9ac8-49f8-99a8-f07f4441100e
2 Likes
Besides: I’ve always been a vinyl collector. So, after decently bemoaning my fate, I’ll delve into what the terms “fingerprint” and “AcoustID” mean. I’ve not the slightest clue.
Thx for introducing me
2 Likes
AcoustIDs should be format independent. An acoustID contains several fingerprints of similar recordings. If you submit a fingerprint through Picard it is compared with similar fingerprints in an existing AcoustID. If it’s very similar it generates the same fingerprint, but depending on format or quality it may have a separate fingerprint, but it will be grouped in the same AcoustID. If it’s not that similar, a new AcoustID is created.
(in theory all fingerprints of the same recording should go to the same AcoustID, but that’s obviously not always true)
If you click on an AcoustID (Recording\Fingerprints tab) you go to https://acoustid.org where you can compare two fingerprints. You can shift the the second fingerprint (compared to the first one) up and down until the right bar turns black (the difference between the first 2 bars):
In this example the fingerprints have no perfect match in any position, but they are similar enough to end up in the same AcoustID.
2 Likes
What I meant with…
If there are multiple acoustIDs linked to a recording, you can try to compare fingerprints of each acoustID and look if they are similar enough to belong to the same recording, because many submitted fingerprints are false submissions of incorrectly tagged files.
Also look at the track lengths of the files which submitted the fingerprints. If they are completely different, they will hardly belong to the same recording. And, what @yindesu pointed out is that different lengths don’t prevent the fingerprints from going to the same acoustID, because only the first 120 seconds are used for the fingerprint.
1 Like
It’s worse than that - 2/3 of my examples are shorter than 120 seconds (2:00).
1 Like
I’ve noticed that too. Shorter fingerprints exist in the same acoustID, although there are some longer than 2:00. Maybe they are accepted as partial submissions and compared with what’s there. (?)
I wanted to add that sometimes they will do this, as @sound.and.vision showed in another thread. ideally, these should also be disambiguated though.
one other thing to look for is conflicting relationships on the recordings in question, say one is a live recording from say, 1989, and the other is a live recording from 1995. I’ve just recently split a recording for similar reasons.
2 Likes
Thank you very much for explaining, but of course this means a lot of work in case of vinyls
Why? There is nothing different with vinyls.
If I go back to your example:
One acoustID is linked to both recordings. The “4:00 recording” has three more acoustIDs. You may like to compare them to the first one:
Click a fingerprint from the first acoustID (both recordings). If there are many fingerprints, pick one with many submissions (although they will all be very similar, because they were grouped in the same acoustID) → copy/paste a fingerprint number from another acoustID into the “compare” field at the bottom of the page → click compare → move up and down to find a “perfect” match…
You will see, that the others are also similar, although one of them has section with no good match, maybe because of very poor audio quality, maybe because of manipulation (remixed?). But usually different versions are clearly different.
And another thing: If you look at the number of submissions, those with little similarity have only 1 and 2 submissions. You can forget about them → merge!
1 Like
Well I guess I have to make a recording from my vinyl to get a fingerprint .-)
3 Likes
Ah, I understand…
Only in case of conflicting acoustIDs (clearly different). Then you should probably submit your acoustID to make sure which one is the correct acoustID for this recording. It’s not necessary in your example.
1 Like
Yes, and these are cases when acoustIDs prove to be most helpful. The 89 recording surely has other acoustIDs than the 95 recording. Of course, if they were already merged, and if you try to untangle the recordings, there’s much work ahead
1 Like