Accuracy of the database?

Tags: #<Tag:0x00007f756fdf4800> #<Tag:0x00007f756fdf45d0>

I’ve been working on restoring my music library, which is quite extensive. In the past, I’ve used TuneUp, which has been pretty spot-on as far as identifying songs yet as TuneUp no longer seems to be in development, I need to look into alternatives, and Musicbrainz seems to be “the” alternative to Gracenote.

Yet, my first pass (using Jaikoz) with MusicBrainz has, to be polite, introduced a significant delta in expected results, to the point where I am now manually tagging over 80% of over 50,000 songs.

In my efforts to determine where the point of failure lies, I’ve decided to focus on a specific subset of music - specifically my Beatles collection, because I (as the kids say) “have the receipts” - I know exactly which albums I do have.

To be clear, the vast majority of my collection also “has the receipts.” I have my CD collection, my husband’s CD collection, my mother’s and my father’s. That comprises the bulk of my collection. However, my Beatles collection - acquired long before the Internet existed, is solely comprised of the Parlophone CDs, 1, and I believe the first Anthology. So if I bang my collection up against Musicbrainz, every song I have should belong to one of those CDs or be a duplicate thereof.

And yet.

A Hard Day’s Night is tagged as I Should Have Known Better (RM1 take 22), A Hard Day’s Night Deluxe Edition, Vol. One, Track 2 of 31,
MB Release ID a8436b07-c2c8-4b19-9ea9-a568c7d9d264,
MB Recording ID efeb57e6-b076-4485-a343-fcf5f5bce7f4
MB Track ID dbc729ba-a3c0-486f-970f-88e31741cc67

This should be A Hard Day’ Night, track 1 from the Parlophone 2009 release. it literally HAS to be, because that’s the only way I acquired the song.

Eight Days A Week - should be on Beatles For Sale, or, at worst the “Red Album” (of which I have no provenance) but Musicbrainz has it as “The US Albums” which was released long after I stopped buying CDs and purchased directly from iTunes.

Eight Days A Week, The U.S. Albums,
MB Release ID 7d00573d-1007-4321-ac9e-e003b714a4c3
MB Recording ID 63dd7ef6-6d6e-44d5-a4d9-190e49223077
MB Track ID b2b6331f-0c67-416b-9f5a-25c4877c2bc4

Sie liebt dich - should be on Past Masters Volume 1, as that’s the CD I purchased. Yet Musicbrainz has this as Sie liebt dich (She Loves You) / I’ll Get You
MB Release ID 395813d4-16c7-4787-b5c0-e9ff0bc864f6
MB Recording ID 7a2287d6-357e-4cb4-adae-f19db3469c8b
MB Track ID 0eee4027-47fc-4993-b389-edb1d82a73a9

I bring these up because not a single song in my collection should be tagged to anything esoteric, and yet my collection is now tagged to albums I never owned, at the expense of albums I do own.

Now, I believe that the music tagger needs to take the blame for not assembling albums and just hitting the first entry and calling it a day, regardless. But the tagger can only work with the data it’s given, and I don’t know if Musicbrainz data is clean enough to accurately tag my collection - or if the ALL the folks tagging have the provenance for their entries. I have “apples to apples” comparisons. Gracenote would absolutely match these songs to the albums I expect. And IIRC, Musicbrainz has unique IDs for each individual file. Which means that someone told Musicbraiz that a song file that should have been attached to, say, Beatles For Sale, or Past Masters, belongs elsewhere and as QA is crowdsourced, it would take decades for the data to be properly attested and the proper albums “rise to the top.”

I could be very wrong with how Musicbrainz works, but right now, based solely on the results I get for the Beatles alone, I do not know if the database is as accurate as people hope.

My concurrent effort is to also work with Jaikoz to see if the tagger can be configured to “build albums” or otherwise prefer the top rated results.

If I am wrong in my assumptions (and I could very well be wrong), any insights into how to get the correct data would be greatly appreciated. I’m going to continue to use my Beatles collection as the “control” until I can make sense of how this system works. I can also provide a csv of the data.

1 Like

I don’t use Jaikoz myself but it seems like this is a problem related to Jaikoz (not developed by the MB team) and not the actual database. I took a quick look and it seems like all Beatles albums that you’re looking for are in the database with the correct release date, etc. There are for example currently two editions of the 2009 reissue of “A Hard Day’s Night” to choose from. Another possibility is perhaps that the files existing tags are messed up or missing, or that there perhaps is some type of setting in Jaikoz that you’ve missed, because matching the (I assume) recording title “A Hard Day’s Night” with “I Should Have Known Better” is very strange behaviour.

You should try the official tagger MusicBrainz Picard. It has settings which let’s you narrow down what type of releases to look for (album or compilations; what type of formats, etc). A simple acoustid scan of the file should be able to at least find the correct recording. A scan of the disc-id from the CD should find one of the 2009 editions (if the disc-id currently is assigned to the correct release). You can also cluster recordings for better matches, and then manually select the correct release. It’s also possible to skip the automated matching process by manually searching for the releases you’re looking for (copy/pasting the URL to the MB release into the search box in Picard also works).

https://picard.musicbrainz.org/

5 Likes

Some of the problem is caused by how matches are made. And I think you may have misunderstood how matches are made, and also how to use Picard to its best.

Picard is a can be VERY weird to get used to. Especially when working with common music like The Beatles as these tracks can be on sooooo many different releases.

The best way to check the accuracy of the actual database is go look at a few examples directly. Do some manual searches in the database. Can you see your actual release there? If yes, then you can use that green TAGGER button to push the data back to Picard.

It is hard to give a direct answer as we need to know some examples. How are you files currently tagged? Which way are you using Picard?

The best way is to hand Picard only small numbers of albums at a time. (I only do one or two at a time) Make sure you use the CLUSTER option to gather tracks into albums BEFORE using the LOOKUP.

It gets more confused if you use SCAN as now it is trying to find matches based on what the track sounds like… and that way lies confusion with compilations.

It is also possible to go into the settings and adjust how it matches - downgrading the matches for Compilations often helps.

Tell us more about exactly what you are doing to use Picard and we’ll help adjust the software to give you much better matches. :slight_smile:

The help pages also point to videos to better show how to make use of Picard. this is not just “pick up and go” software, but once you get your head around it you will find it VERY powerful.

I have no idea about Jaikoz.

2 Likes

In the case of Eight Days a Week , the recording you reference is on Beatles For Sale, as well as on The U.S. Albums, and the 1 compilation, among others. That recording “belongs to” all those CDs (and their LP, cassette, and digital equivalents). So in this case jaikoz has found the right recording, but whatever heuristic it’s using to pick the specific release may be off.

The Hard Day’s Night / Should Have Known Better mixup seems to be of a different sort. Sometimes a problem like this is caused by an incorrectly attached acoustid, but in this case the recording in question doesn’t have any acoustids at all. So unless your file is already tagged with the incorrect MB recording ID, I can’t think why it would mismatch that way.

Have you tried the Jaikoz forums? As @jolietjake and @ivandobsky suggest, there’s more expertise here with MB Picard than most other tagging applications, but I know there are lots of folks using Jaikoz with MB.

2 Likes

I’ve reached out to Jaikoz, and we’re doing some triage on the behavior, using The Beatles files as the control set. This should inform, at least, Jaikoz’s behavior in relation to the data. My theory being that the issues I am seeing is a compound of unfamiliarity with Musicbrainz and Jaikoz and possibly expecting a specific logic that may have been unique to TuneUp.

My expectation is that as we start pinpointing specific behavior, I have to just add my data for consideration. It will take a long time. I have a lot of song files…

In any event, I appreciate the feedback.

1 Like

I’ve tried Picard, but it is apparently having permission issues with Catalina and won’t write the changes. But that is a battle for another day. I’ve reached out to Jaikoz for assistance as well. Once I figure out how to allow Picard to write changes in a Catalina world, I’ll give that a spin and update on my findings.

In a Jaikoz world, based on what I’m reading, is that it may be choosing the first option available and it does not really offer alternates. I see this mostly with genres, where Jaikoz will pick the first listed genre, even if it has been “downvoted” into oblivion by Musicbrainz, while higher rated genres are ignored, but Jaikoz admits that this is a known behavior and I have a workaround. It makes me wonder if the app is doing the same with how a song with a single ID is associated with multiple albums.

It is also very likely that there is functionality specific to TuneUp and how it interacts with Gracenotes I am accustomed to that either does not have an implemented analog in Picard/Musicbrainz/Jaikoz or cannot be implemented and I will have to set my expectations accordingly.

Thanks! I am fighting Catalina’s security with Picard right now - Picard won’t write the changes it identifies … but that’s on me to figure out. That’s why I can really only discuss Jaikoz - but I also asked the same question on that forum.

Once I get Catalina to behave with Picard, I’ll update with results.

That is the same with any application that is on the Apple Mac and not in the Apple Store. Just go into the preferences and give it permission. (If you don’t know where this is on your Mac I may be able to dig out the thread on this somewhere)

Fire up Picard and then go back to the Apple Finder. Open SYSTEM PREFERENCES and locate SECURITY & PRIVACY. You’ll need to “unlock” this with your system password. Then in the bottom half of the screen click on OPEN ANYWAY to give permission to Picard as a trusted application on the Computer. That should then work. (I think it is developed on a Mac, so there isn’t anything fundamental stopping it) See below… Picard now signed so none of this needed…

I have no idea about Jakeoz. I used to use Winamp and remember Gracenote. I thought that shut down? I think Gracenote used some kinda serial number off of the CD to do the lookup. The main difference with Picard is you place your CD into the drive and do a CD Lookup. It then uses the actual layout of the CD in the drive to get a match. The CD’s “Table of Contents” with is lengths and offsets. This is fairly well unique to each Release, but will often be mimic’ed with represses.

I think some of your confusions will be caused by Picard going into deeper details with the albums. Far more different variants will be in the MusicBrainz database than was in Gracenote. It is also coming from a different angle as to how it works.

Have a look at that link to the help pages and follow the Quick Start example on: https://picard.musicbrainz.org/quick-start/

1 Like

This is all not necessary, Picard is properly signed and will run without the need to override security settings.

If I understand @araquen correctly the issue is that changes to the files are not saved. Can you give more details what is happening? Some general advise:

  • Please make sure you have enabled “Save tags”, otherwise nothing will get saved.
  • If there is an error saving the files this will be indicated by a red icon. A common cause is missing write permissions on the files, in this case give the files proper permission with Finder. You can also find more details in the error if you open Help > View Error/Debug Log in the menu.

Feel free to open a new thread in the Picard category if you need additional support using Picard.

4 Likes