Different angle on this, and it may or may not fit your setup, but it changed things for me.

I’ve been accumulating audio since the early 2000s. Twenty-some years of ripped CDs, vinyl rips from release groups, random downloads. Several thousand albums. At some point I accepted that tags were never going to be consistent across that pile, because the sources never agreed on anything in the first place: a 1979 vinyl rip tagged by whoever released it, a CD rip from 2004, and a modern FLAC have nothing in common metadata-wise, and no tagger reconciles that without me babysitting it.

So I stopped trying to fix the tags and leaned on the folder structure instead. My tree already carried everything I cared about: genre at the top, then artist, then Artist - Year - Album, with subfolders for live, compilations and singles. That structure took years to build and it’s reliable in a way my tags never were.

I run Velvet (mStream fork) as the server for it. It serves the library by folder structure rather than by tag database, so there’s nothing to scan, nothing to match, nothing to get wrong. And I vibe-coded Sharesonic, an Android client for it, because I wanted to browse my library the way it’s actually organized and share a track, a playlist or an entire folder as a public link without exporting anything. Personal project, it’s on GitHub if anyone wants it.

To be clear about what this does and doesn’t solve: it doesn’t repair your files. If Picard truncated audio on you, that’s a real problem and it needs fixing at the file level, not at the player level. Worth checking whether the audio itself is actually shorter or whether it’s just the duration metadata disagreeing with the stream, because both look identical from the player side and only one of them is destructive.

The trade-off is real. You lose cross-cutting queries: no “everything from 1977 across all artists”, no proper album artist handling for compilations. If that matters, stay on Jellyfin and keep grinding through Picard. If you mostly navigate by “I want to hear that specific album”, folders are faster than any tag database.

I’m working on a script that checks my files, dir structure, and audio metadata for issues. For basic problems that can be fixed with a command, the script outputs a suggested command in a csv. I’ve cleaned up many issues with the script so far…

Been thinking of sharing it but it’s still very much a WIP.

I use a manual approach for my music:

First, I use Musicbrainz Picard to tag my music according to their database. Most of the time, “lookup” just works, bacuse the source files are already pre-tagged sufficiently, sometimes I need to “analyze” the files, which is slower, but works with whatever “lookup” didn’t catch.

Then I use kid3 to manually clean up my tags, like fixing Genres. I also use its renaming function to keep the filenames uniform - Artist/YYYY - Album/## - Title.ext

After that, the files get normalized using replaygain/mp3gain.

Only then will the files get moved to my NAS, where every device will be able to access them, and overnight my phone will automatically grab any newly added music and copy it to the phones SD card.

Yeah, at some point, I told Picard to just re-sort everything and that was not a good idea. Some songs got renamed or put under a different artist, because I guess, the actual song is not in the MusicBrainz database.
I do have a backup that I should restore actually, but yeah, difficult to notice such changes and especially to know how much damage was done.

I did also figure, I should just add the unknown songs into MusicBrainz, but boy, I added one album and that took more than half an hour…

Definitely leaning more and more towards that unsorted life…

beets with a whole bunch of custom scripts. I tried using picard but found it to be completely useless, it rarely finds a match and when it does it decides it belongs to multiple album editions rather than just 1.

I use Picard, I never had a issue of it changing the actual file, just adding mp3 tags and cover images. I didn’t even know it could mess with the length of the song, maybe check your settings? Most issues I’ve had with Picard have been connection issues, but that usually gets fixed with clicking the refresh thingy.

Anyway, I spent like a week doing 4-8 hours of “work” per day to get my existing music library tags and filenames sorted with Picard. Since then, all new music that gets added to the PC is put into a temp folder, ran through Picard and only then moved into the music library folder.

For the file tags I use the defaults that Picard comes with, I also have a plugin that adds additional musician information into the song. Picard also changes the directory and filenames into /Artist Name/Album Year - Album Name/01. Title.flac, so for example /Björk/1993 - Debut/01. Human Behaviour.flac

Recently I started adding lyrics as .lrc files, but that looks like its going to be more work than its worth. If I wanted all songs to have lyrics files, I’d have to go through most of my non-English language songs manually, the apps seem to have some issues with that. Also, for stuff like Kent, who have 2 albums in Swedish and English, the apps don’t seem to understand that and insist on using the English lyrics for the Swedish versions.

I just started properly tagging new music at some point and now my library is largely tagged correctly.

For tagging I use kid3, great KDE application for that.

I also sort my music into a folder layout like this to know where I got it from and potentially upgrade the quality.

MusicRaw
  - Artist1
    - Bandcamp
      - Album1
    - Tidal
      - Album2
      - Album3
    - Youtube
      - ytrip album
  - Artist2
    - Album1 (not moved to a separate source)
    - Tidal
      - Album2
      - Album3
    - ytrip album

I then have a script to loudness-normalize (come fight me) all music and write it to another subfolder.
A second script takes those normalized files and symlinks them into a structure like this:

Music
  - Artist1
      - Album1
      - Album2
      - Album3
      - ytrip album
  - Artist2
      - Album1
      - Album2
      - ytrip album

I tried switching to Picard, but for somewhat obvious reasons it doesn’t support my insane structure.

i get flac files from legal and illegal sources. convert them to opus. one copy stays on the pc, one copy stays on the phone. both are played with foobar.

I’ve gone back and forth on this, but ultimately, I enjoy the tagging process and accept it as part of my workflow.

When i’m adding a release to my collection, I throw it into a temo directory, open the album folder in MP3Tag, query MusicBrainz (or Discogs if necessary) for the basics, replace the cover image if I’m not happy with the one from the source, add additional genre tags and do cleanup, auto renumber tracks/discs. Once it all looks good, I use the tags-to-filename feature to rename everything to AlbumArtist/[OriginalYear] Album/### - Title (### - Artist - Title for multi artist releases) then copy it to my library.

It sounds like a lot, but it isn’t bad. Like I said, I enjoy it.

Updating tags shouldn’t change the actual music though. That seems really strange and like something else is going on.

By Artist and Album, then let Plex handle the rest. It’s pretty much the standard format for torrents so 95% is just drop in and done. I occasionally do youtube DLs for very recent releases but those I rename and sort by hand.

Alphabetically.

Usually beats is the solution but it uses the same source as picard. Songs shouldn’t get shorter though? Sorting should be the same though, independent of the software.

Music/artist/release/title

I gave up a couple of years ago. It is too much work for too little gain.