The concept of AcoustID

[ernstlx]

Hi,
I still don’t understand which recordings should be (correctly) linked to a specific AcoustID:

1. specific recording including correct title - yes, of course
2. correct recording (content of file), but tagged wrongly - I suppose no
3. different but similar recordings (?)

I’ve got a very simple example. No bad data, only technical issues. Because of switching track positions on the Japanese release files were tagged with wrong titles (recordings) from the French or US release.

“3 Deuces” (Japanese #5, F/US #1) switched position with “Burning Down the House”. Here is the AcoustID of #5 of the Japanese release: (After comparing the pattern of JP and F/US submissions, I disabled the wrong MBID)

(During the first seconds the F/US track 1 is differently mixed (applause/audience), then it’s perfectly the same)
That’s case 2 … okay so far?

On the F/US release side, I’ve got case 3:

The first 3 Deuces is the Japanese version. You can distinguish them by the live disambiguation comment of the F/US recording. (the third is from a compilation and should probably be merged)

Should the Japanese version be disabled as well?
Is it a good thing to have more than one recording associated to an AcoustID? Is that helpful? Is it damaging?
I hope for answers.

Recordings from example:

3 Deuces

3 Deuces

(third: Recording “3 Deuces” by Marcus Miller - MusicBrainz)

[yindesu]

An AcoustID with more than 1 listed fingerprint is usually supposed to be linked to more than one recording. See my example:

Merging DJ-mix tracks with same fingerprint as normal one

Please never merge DJ-mixed tracks with the original/separate tracks. ISRCs mean nothing. In compilations of edited/cut/DJ-mixed tracks, the ISRC used is usually for the original, pre-edited recording. So the same ISRC will be present across multiple MusicBrainz Recordings. AcoustID intentionally ignores a variable amount of the early part of the fingerprint, allowing entirely different edits to have the same AcoustID - see the comparison. The minimum definition of a DJ-mix is that the begi…

[yindesu]

Here’s another example of an AcoustID that is correctly linked to a dozen recordings:
https://acoustid.org/track/cd048052-895a-4abf-98ff-d4cfc4a04e9a

Same backing track, different vocals.

[ernstlx]

So similar recordings should be linked to the AcoustID submitted from another recording?
Okay, then I should only disable those linked to a not-similar recording (wrong track or something) - live and studio?

EDIT: in the second example they are all same length - are they really all different DJ-mixes? (the comparison shows a difference pattern similar to my example…)

[yindesu]

Track "cd048052-895a-4abf-98ff-d4cfc4a04e9a" | AcoustID are different recordings/mixes - nothing to do with DJ-mixes. I am providing an example of how AcoustID is intentionally loose, which makes it difficult to decide whether an AcoustID-MBID link is correct if you do not hold copies of all of the files.

[ernstlx]

I withdraw the question about DJ-mixes. (thought about what a DJ-mix is)

@yindesu

how AcoustID is intentionally loose, which makes it difficult to decide whether an AcoustID-MBID link is correct if you do not hold copies of all of the files

… therefore we keep a bunch of AcoustIDs with countless pattern for each recording…

(Thanks!)

[IvanDobsky]

I would have expected the crowd noise at the start of a Recording to cause a different AcoustID. This is why you can get two recordings that are linked to two different length AcoustIDs. Recordings can get merged when there is an early fade, but would lead to multiple AcoustIDs listed to that Recording.

I would expect 20 seconds of different to cause different AcoustIDs even if the Recordings are the same.

Now this can lead to confusions on the AcoustID screen if data was submitted from the 6:23 edit, but then the recording was merged to the 6:42 edit. It means you then see the “wrong” length shown as the recording has multiple lengths.

I don’t think that is happening in this example. The different length recordings are still separate. So they should have separate length AcoustIDs

Do you have copies of either or both of these albums? What numbers are you getting? Which length edits do you have?

It is very common for errors to be uploaded. I am not surprised when I see mix-ups like this. Some people will use Tagging apps to tag a re-issued CD with data from the original vinyl to get that release date. They then hit “submit acoustIDs” not noticing they have totally different length edits of some tracks. This is one way bad length AcoustIDs get attached to the wrong versions.

Personally, in this example, I would chop the “wrong” length recordings away. Mainly because the recordings are still kept very separate. But then I am also getting pretty brutal lately with duff data in AcoustIDs.

(I will look at this example a lot closer now and re-write the above if I am talking rubbish)

[IvanDobsky]

@ernstlx

2. correct recording (content of file), but tagged wrongly - I suppose no

This one confuses me. If you are seeing a bad track list on an album, then that track list can be corrected according to guidelines.

Even when the track list is bad and kept bad because someone likes typos - then the Recording should always be corrected to the actual recording name.

Therefore any AcoustID page should always only be listing correct Recording names.

Yes, I have often used AcoustID to spot and correct track listing errors. But once the error is corrected on the track list, this then will show up with the correct Recording name on the AcoustID page.

[yindesu]

Let’s also be clear - we do not control how many patterns are listed on an AcoustID. That decision is made by the AcoustID server.

Unlinking an MBID will not drop the fingerprint from the AcoustID.

[ernstlx]

@IvanDobsky

This one confuses me. If you are seeing a bad track list on an album, then that track list can be corrected according to guidelines.

But there is no bad tracklist associated: the recording is linked to is the correct recording at the correct position for #5 Burning Down the House (F/US)
Click it:
https://acoustid.org/track/a90059cb-358d-4a37-9a49-7e860775ec77

[IvanDobsky]

@ernstlx

But there is no bad tracklist associated: the recording is linked to is the correct recording at the correct position for #5 Burning Down the House (F/US)

@ernstlx

Because of switching track positions on the Japanese release files were tagged with wrong titles (recordings) from the French or US release.

Sorry - this is why I am getting a little confused. Too many tangents in the questions at the same time. Lets ignore that side then.

-=-=-

So, if we go back to what we are seeing on those two albums. In your 3 Deuces example above I would disable the 6:42 recording. I expect that is someone selecting the wrong version of the album when they submitted their data. 48 other people picked the right version with the 6:23 edit.

Those 20 second of audio difference woudl cause a different acoustID - but they would have overlapping fingerprints.

Notice when you click on that 6:42 Recording it also has a correct length AcoustID also linked with it giving you more confidence in removing its link to the 6:23 acoustID.

[ernstlx]

@IvanDobsky

Those 20 second of audio difference woudl cause a different acoustID - but they would have overlapping fingerprints.

The 20 extra seconds are probably at the end (after 2:00).

@IvanDobsky

I expect that is someone selecting the wrong version of the album when they submitted their data.

Probably because this was the closest match to the correct track - the Japanese release was added quite recently.

[IvanDobsky]

@ernstlx

The 20 extra seconds are probably at the end (after 2:00).

I would still have expect a different AcoustID - and this seems to show up by other people who have submitted correct AcoustIDs for that 6:23 version.

@ernstlx

Probably because this was the closest match to the correct track - the Japanese release was added quite recently.

Well, yeah. That is a good reason. Also many people really don’t care about accurate matches. Or plain didn’t notice that they had different length recordings.

Lets look at your recording of 3 Deuces. I assume you have one of these albums. Do your AcoustIDs match up as you would expect when looking at this release?

(Sorry - I have spent far too many hours in AcoustIDs lately so am more than aware of the percentage of cruddy data in there… so know to be very suspicious of anything looking odd like this which has only a single sample linked)

Happy to help focus on specific examples and tell you what I’d do with them.

[ernstlx]

@IvanDobsky

so know to be very suspicious of anything looking odd like this which has only a single sample linked

you can have this if you work on jazz albums

@IvanDobsky

Also many people really don’t care about accurate matches. Or plain didn’t notice that they had different length recordings.

If you don’t like to add your own release and you reorder your tracklist instead (for matching the F/US tracklist), you get at least a nearly correct MBID for your recordings…

[yindesu]

AcoustID only fingerprints the first 2 minutes.

[IvanDobsky]

Okay, as an example, I’ll pile in on those Releases and suggest what I’d be doing.

The nice thing here is these releases seem to only have one or two AcoustIDs per track. Easier to clean up.
Example: Recording “People Make the World Go 'Round” by Marcus Miller - MusicBrainz

That track clearly has a length of 11:09, but one AcoustID shows as 1:27. I’d disable that one.

Next: Recording “When Your Life Was Low” by Marcus Miller - MusicBrainz
Next: Recording “Panther” by Marcus Miller - MusicBrainz
Next: https://musicbrainz.org/recording/f2022b60-6b5b-4874-952a-b1eb50cb3005
All have two different samples that are close in length. I’d leave these as not an obvious error.

next: Recording “Power” by Marcus Miller - MusicBrainz
5 mins acoustId on a 6 mins track - too much of a different. I would unlink this one. Especially due to the lack of other data.

finally: Recording “Intro” by Big Doug Epting - MusicBrainz
Again, a pair of close AcoustIDs. So these stay in place.

Ah-ha - I see now. The similar lengths are from the Japanese edit of the Release. These recordings are close enough that they have been merged, so now the two slightly different AcoustIDs are on the same recordings. That seems logical as the recording is the same, just slight differences in amount of audience added.

Don’t ask me why I went through the tracks backwards. Just did.

I can’t see any major issues in there. I see the Japanese edition has a lot of differing lengths and some extra tracks. I understand why those tracks have been megered in most cases - corrrectly leading to two different length AcoustIDs pointing to one Recording.

Those tracks unique to the Japanese release should have the “wrong” length disconnected.

I would delete any of those that didn’t have other meta data listed that don’t match and have times a long way out.

Sorry - I kept editing this as I looked deeper…

@ernstlx

you can have this if you work on jazz albums

Seriously - this jazz album is easy. You should see what old Floyd bootlegs get like. Especiallly with the cross linking to other gigs.

@ernstlx

If you don’t like to add your own release and you reorder your tracklist instead (for matching the F/US tracklist), you get at least a nearly correct MBID for your recordings…

Don’t understand this comment. Sorry. Maybe you misunderstood my meaning. I was trying to give a reasoning why a lot of bad data appears in MB. (We seem to be going down too many tangents)

[ernstlx]

@IvanDobsky

Example: Recording “People Make the World Go 'Round” by Marcus Miller - MusicBrainz

That track clearly has a length of 11:09, but one AcoustID shows as 1:27. I’d disable that one.

@IvanDobsky

Example: Recording “People Make the World Go 'Round” by Marcus Miller - MusicBrainz

That track clearly has a length of 11:09, but one AcoustID shows as 1:27. I’d disable that one.

OK

@IvanDobsky

next: Recording “Power” by Marcus Miller - MusicBrainz
5 mins acoustId on a 6 mins track - too much of a different. I would unlink this one. Especially due to the lack of other data.

But how is it possible, that there is a 5:01 pattern, if the only submitting recording is 6:02?
…

@IvanDobsky

Seriously - this jazz album is easy. You should see what old Floyd bootlegs get like.

I’ve seen AcoustIDs for Peter Gabriel and I suppose for Pink Floyd it’s much worse

@IvanDobsky

Don’t understand this comment.

I thought about how one can manage to tag his file with a recording at the wrong position (forget it)

[IvanDobsky]

@ernstlx

But how is it possible, that there is a 5:01 pattern, if the only submitting recording is 6:02?

Because people submit carp data. Notice with this one there is only a single sample uploaded, and no other meta data. Also notice the fingerprint does not match at all with the other AcoustID on that recording: Compare fingerprints #11832608 and #36053293 | AcoustID Therefore

Also notice how all the other examples of double AcoustIDs have good metadata at the bottom half of the screen to backup their sources.

@ernstlx

I’ve seen AcoustIDs for Peter Gabriel and I suppose for Pink Floyd it’s much worse

Bigger the band, the bigger the mess. The more bootlegs that repeat track names, the bigger the mess.

Sometimes you can find the same live album associated to a dozen OTHER versions of the same gig. Always lined up “by name”. They stand out as usually there is only a single “source” uploaded.

*(Note I re-wrote my post above after looking at the Japanese release)

[ernstlx]

@IvanDobsky

the fingerprint does not match at all with the other AcoustID on that recording: Compare fingerprints #11832608 and #36053293 | AcoustID

No, not within the 2:00 mark.

@IvanDobsky

Also notice how all the other examples of double AcoustIDs have good metadata at the bottom half of the screen to backup their sources.

Why is there no submitted metadata on this AcoustID?
https://acoustid.org/track/77832f1b-0760-4a8d-8398-872626a0c60f

[IvanDobsky]

@ernstlx

Why is there no submitted metadata on this AcoustID?

Look in that image. A single source, two fingerprints. Tells me only one person ever submitted. And nothing at bottom half of page - no “metadata”. Tells me not to trust this.

I now click on the Recording and look a the other AcoustID

Now this one I trust. Over a hundred matching fingerprints, another 50 people on similar fingerprints.

Over 50 sources linking to AcoustID via Musicbrainz

And a heap of metadata from other sources.

This second AcoustID is one to trust. The first one is junk.