I tried to see a pattern, but I see that some albums which were originally bought digitally and had all their metadata in the files from the day I bought them show up here as unidentified even though all the track titles are still pulled from the files with no issue.
I picked one up at random and tried to manually identify it, but I couldn’t even find it manually using the correct artist and title, either using the English titles from iTunes or the Japanese titles on the CD itself.
The “don’t lose hope” dialog assures me that newer albums take some time to show up but this one is not exactly new.
So I am wondering, exactly what does “Unidentified” mean vs. an album being identified? Is it worth enduring this terrible user experience to try and get over 500 albums identified manually?
It sort of depends on the cause for not being identified, of course. If it is very obscure and not in any database, that’s a problem. But you have so many it’s probably not that. If you posted some details about an album that you think typifies your issue, it could help get advice for you.
There are programs like SongKong that will go through your library and identify tracks/albums based on the existing metadata. It will also fill in gaps to help Roon identify it. The developer @paultaylor is active on the Roon Community. You will find various threads discussing how it is used.
There are other programs, of course. But that’s the one that comes to mind with your issue.
This one is now identified because I went through the manual process to identify it. But I was baffled as to how it didn’t get identiified automatically still, because it was the first title in the suggested list when I manually searched.
So I assume that some number of them are going to be like this - I will manually search, it will find it, and then I will just curse that it can’t do it automatically for me.
While others will be like the first one - not in the database at all, even though it was available through a major digital music store and thus not really lacking in metadata.
I’ve found MANY, MANY albums in my large collection (20K+ albums) that remained unidentified simply because the length of the ripped files varied by a few seconds from the length specified by Roon’s info provider, often differing by only one file. All the track titles were correct and yet… Roon still puked. A little more sophisticated AI or fuzzier logic would be appreciated. Or even just the option to ignore track timings as an ID criterion.
This is spot on, I think the basic roon algorithm looks at the track lengths of your files and then looks up in its database for an album with a track listing with the same tracks lengths in the same order. If the album is in one of its databases then this can find a match, and existing metadata in the files can be used to determine the right choice on those occasions where more than album is found. But if the track lengths on the user files don’t match up withe the database then roon has to rely on existing metadata only. If this metadata includes MusicBrainz Ids then it easy for it to use that as a match but if the only metadata is items like title, album and artist then it can be much harder to find an acceptable match even if the album is in one of Roons databases.
If the album is using a non latin script as in Japanese releases then metadata matching is much harder because the rules for matches text are different, and I suspect this is harder for the roon team to deal with. Also the databases that roon uses may only have the English or the Japanse version, and the user may have the other version.
Having to use another application such as SongKong may not be ideal, but it is a pragmatic way to get more matches. Because it uses different algorithms to Roon, it can find matches that Roon cannot and if it is a MusicBrainz match Roon can then make use of the MusicBrainzIds it adds. SongKong can also search Discogs and Roon can make use of the metadata added from Discogs.