||[Oct. 15th, 2006|09:33 pm]
I downloaded the MusicBrainz database (316 MB compressed), and plan to write a non-fingerprinting auto-tagger on the flight.
-- directory hierarchy (including dates for albums)
-- track titles (including often the track positon)
-- song duration, in seconds
I bet I can write a really damn good auto-tagger without even resorting to fingerprinting, which doesn't seem to work too well. (At leats iEatBrainz seems to kinda suck....)
There's just so many clues to use from the filesystem alone.
Please tell me somebody's done this?
Considering that CDDB et al basically work from just a list of song lengths, I think you shouldn't have much trouble making it work nearly 100%. I think people rely on fingerprinting for things where you just have the song, or you have a pile of unnamed / poorly named songs. It's a giant win if they're already in a hierarchy so you only have the mp3s from a single song in a given directory.
I'm not sure what data is in the musicbrainz database; the freedb database can be downloaded and is 480MB compressed.
I feel like EasyTag and The Godfather might be able do this, but neither of those are Mac apps, unfortunately. Re: iEatBrainz. I had some pretty good success with it in the past, but now it seems not to work at tagging files which it had previously tagged just fine. It's possible that an iTunes/QuickTime upgrade broke it?
2006-10-16 06:08 am (UTC)
iEatBrainz doesn't go through iTunes/QuickTime, not that I can see.
Thanks for the EasyTag reference! I'll try that out first.
I was pretty sure that iEatBrainz used QuickTime to to decode the audio before fingerprinting. From January, in the Developer's Blog
iEatBrainz is a bit beyond resurrecting code wise (It does still work, but changes by Apple or MusicBrainz will certainly break someday in the future), while parts of it could be harvested, I’d suggest to anyone thinking about it to strongly consider a cocoa gui for Picard as it has a lot more features and with MusicBrainz fingerprinting up in the air it’s a much safer approach.
I've had problems with EasyTag randomly trimming the title field. The UI is kinda crappy, so you don't always catch the trimming, since it leaves the filename intact. Just a heads-up.
Semi on-topic, something you might want to know. (Helps you in the future, but unfortunately not in the preset.)
iEatBrainz only uses the old TRM fingerprinting format, while MusicBrainz is moving to PUIDs as of Picard 0.7 and 0.7.1
You probably knew this already.
Picard is being re-written using Qt for ultimate cross portability (and reliability, silly wxWidgets).
Someone in MusicBrainz IRC mentioned creating a userspace file system using FUSE that would create a directory-detached (...perhaps agnostic is the better world I'm looking for?) organization of music files.
Something that would allow accessing a file, regardless of filesystem location, by accessing it's metadata (artist, title, album, year, whatever).
It was mentioned, but I don't believe implementation went underway or anything.
2006-10-16 07:10 pm (UTC)
A virtual filesystem for music is something that's been on my “I'll write that one day…” list for years. However, I'm not entirely convinced that the kernel filesystem layer is the right place to do it. It seems better placed in the shell's virtual filesystem, where it can do more clever stuff like presenting the right icons for albums and stuff like that.
This does limit you to using apps that support your shell's virtual filesystem layer, of course; I wonder if there's some way a compromise can be achieved so that all of the important stuff is in the kernel-level FS (which is actually FUSE-based, but the apps don't need to know that) but the fancy-shmancy stuff can be layered on top at the shell level. I don't know nearly enough about how KDE and Gnome's virtual filesystem layers work to speculate on this… more research to be done.
2006-10-16 09:31 am (UTC)
Is the problem you're trying to solve that the id3 tags in your mp3s do not correspond to their file names?
2006-10-16 03:51 pm (UTC)
It's that I often don't have id3 tags, but have everything organized in directories correctly, and named correctly. So all the information is there (in one of 3 or 4 similar forms), just not in the id3.
2006-10-16 05:03 pm (UTC)
Are your MP3s from, like, the nineteen-hundreds or something? Did you rip them with a steam turbine?
Have you found a decent library to query CDDB or other music metadata DB? I just ran into this problem where I need an app to run through all my MP3s/OGGs/AACs, properly tag them, and then group them by source directory for de-duping.
I couldn't find any decent Java libs to do the querying although maybe perl has better support.