I’m working through a music collection I “acquired” when my wife and I started living together many years ago. There is a lot of stuff with questionable origin and even worse tagging. In theory Picard + AcoustID is perfect for this. In reality - I’m struggling. I quickly learned that I can’t just trust Picard’s match based on the fingerprint as there are often multiple matches and it does not always get the best one. So, I’ve started doing a manual “Scan” and then right click on the AcoustID and “Lookup in Browser”. But, I quite often get results like this.
Two questions:
- any tips for analysing that track to determine the best match?
- is there anything I can do to improve that data?
Which tags are you interested in being correct? All of them, or just Artist, Title, and Album?
Just Artist, Title and Album - that will allow me to find everything else.
Since posting I’ve realised that this was not a good example to use - I was a bit overwhelmed by the rows and rows of different user-submitted metadata and missed the 105 sources pointing to one particular recording.
My general case question still stands though - any tips, beyond looking for a large number of corroborating sources, to deal with “busy” data like this? And, is there anything I can do to improve it?
I don’t know if others will agree with me, but:
- Make a copy of your music files
- Add the music folder to Picard
- The files should appear in the right pane
- When Picard finishes, click Save
Another method is:
-
Make a second copy
-
Select this option:
-
Add folder
-
The files should appear in the left pane
-
Click Lookup, and if anything remains, Scan
-
Save
Compare the two resulting folders. This is, of course, an automatic method. It’s not perfect, but it’s faster than the manual method.
I don’t think that will work. Most of my library is correctly tagged - I’m working through the 5% that’s a mess.
Here’s a better example: Track "30a686b3-fdb4-4b86-a67e-1f9d6b3167d3" | AcoustID
I have the album, and I have this questionable random mp3. The AcoustID track page has 31 sources telling me it’s the album version but I’ve listened to it and it very obviously isn’t.
I can only add that the preferred method on the forum is: Cluster, Lookup, Scan.
And how many files is your 5%?
2,500 - and they’re mixed in with the good stuff
Do you have studio albums, compilations, or your own The Best?
I’m not sure what you’re asking there. It’s stuff collected by two people over the last 40 years - there’s all sorts of stuff.
Or maybe this.
Hardcore 
Separate those 2,500 files, clear the tags, and run Scan.
Of course, everything on a separate copy.
You have nothing to lose.
No, you can’t fully trust Picard’s match based on the fingerprint. If your files contain reasonably accurate metadata, “Lookup” (comparing metadata) will lead to better results than “Scan” (comparing acoustic fingerprints).
If you do “Lookup in Browser” from the left side (untagged files, which you should first group to “clusters”), you will get a list of possible releases from which you can choose the correct one and “Open in Tagger” (Picard) from there.
For the first recording, a fingerprint matching this acoustID was found 31 times, but if you can falsify it, you should disable the studio recording and it will no longer confuse Picard’s scan results.
The studio recording has a list of acoustIDs associated and many of these acoustIDs are probably wrong. This acoustID is most likely correct for the studio recording. Matching fingerprints were submitted many thousand times and additional metadata also points to the studio recording. If you compare fingerprints of this acoustID with the one with 31 submissions (yours), you will find no match at all: Compare fingerprints #21037807 and #13934970 | AcoustID
The live version is not well documented and you can only improve data quality by submitting correct fingerprints for a recording you found by other means. (only submit fingerprints if you’re sure about it)
If your file do not come from a “Rarities, Acoustics and Live” bootleg compilation, the matching release was possibly not entered yet. But as a starting point, I would look at this release and its recordings.
If you search for recording title and artist, you will find other live recordings of this track, possibly with similar acoustIDs associated.
You may try to search for them with Scan by adjusting “Preferred Releases” in Picard’s options - this one is certainly live, therefore Live to 100. But it will still fail if there was no similar fingerprint submitted.
4 Likes
Lukáš Lalinský will be offended.
I don’t think so. Scan quality depends on fingerprints submitted for recordings and if that’s a mess, there’s nothing a developer can do.
3 Likes
The idea was great. But at some point, things started to go wrong. You know…
+1 that cluster > lookup is the go-to workflow for albums.
Save scan for really tricky situations/derped tags.
3 Likes
But you can’t deny it.
AcoustID and Scan are still powerful and useful tools. In my 16,000 files, I only have 1% errors.
Of course, I don’t consider this case:
10. Leonard Cohen - Suzanne.mp3
and
10. Leonard Cohen - Suzanne (radio edit).mp3
to be an error.
Agree. But what’s causing me grief is the random bits and pieces. Like single mp3s with names that don’t match the tagging, that I like and want to keep - that I’m trying to properly identify so I can try to track down the actual album, or whatever they originally came from.
1 Like
But, I do consider that an error. And with all possible respect, because I don’t know you at all - I really hope you’re not uploading to AcoustID.
3 Likes
I see the artist and the song title. That’s the most important thing. The rest can be filtered out.
