SongKong Jaikoz

SongKong and Jaikoz Music Tagger Community Forum

Matching difficulty

I have the following album - https://musicbrainz.org/release/e331893f-a5ff-4713-8c13-11ce193f7208 . The tags were a bit messy, but neither SongKong nor Jaikoz could match it to MusicBrainz (at any level). I changed the album tag to match exactly and still no luck. I tried Picard and it matched 100% with no issues. So I saved that and matched again in SongKong/Jaikoz using the MusicBrainz IDs to match - of course no problems now! Any reason for this? (BTW the album was added to MB two years ago - it shows as only updated recently because I added the cover art).

If you start SongKong, run Help:Create Support FIles and send me the zip file created I can look into it. I cant tell with just the information provided because it depends on your audio files as well as the MusicBrainz release you are trying to match to.

But with Jaikoz, if you do know the release you want to match to you can use Action:Remote Correct:Match to Album:Match to Specified MusicBrainz Id and if possibly it can it will update form the MusicBrainz Id you have specified.

I ran a test of SongKong against a folder with 3228 songs. 2460 were matched to a MB release (76%) which I consider not bad considering my slightly oddball collection. There were 56 albums which were not matched to releases. 14 of these actually had MB releases, of which the album I quoted was one. I then checked another of the 14 - https://musicbrainz.org/release/06aa2341-81fb-4a8a-821f-6b6131b0d1e5. This only matched in Picard after changing the Album and Artist to match MB. Jaikoz could not match it even with the “correct” Album and Artist. However, your fall-back suggestion worked and is quite an easy solution*. I’ll run SongKong against another one of the 14 and send the results.

*BTW for some reason the composer didn’t get added to the album name, despite having this option checked in the preferences. I tried re-running remote correct, but still no composer in album.

I don’t think 76% for Classical is too shabby, since the Classical coverage of all the major database such as MusicBrainz and Discogs is not as good as it is for Pop/Rock

Note the 76% is when SongKong could completely match a group of songs to a release. In the summary you should also have listed:

a (X%) songs matched to MusicBrainz song only
b (Y%) songs matched to Acoustid song only
c (Z%) songs matched to Discogs

The values of X and Y can be added to the 76% of songs matched to get total percentage songs identified, but for MusicBrainz Song Only SongKong acted cautiously and did not update album fields in case multiple songs in same grouping that appeared to be in same release were split, you can remove this restriction through the options.

For Acoustid Song Only the songs were identified in Acoustid database with basic metadtata (artist, track) so useful for tracks with no metadata, but no match could be found to a song in MusicBrainz.

For Discogs there are probably some releases that we could not match to MusicBrainz but we could match to Disocgs.

The songs fixed by X and Y often can get get matched to a MusicBrainz release if you rerun SongKong, since this additional metadata may be enough to find a good match. But unfortunately you cant test this out with the trial because the changes from the first run wont be saved.

Jaikoz doesn’t currently do Acoustid Song Only or MusicBrainz Song Only matches but planned to be added.

I would have to look at your support files to see why Composer did not get added.

Am sending you an email with the zipped support files Report 20 is for this one https://musicbrainz.org/release/03096267-3b3d-4eed-9e7d-87ac3bc2305a. Picard matched 6 tracks (of 19) on first attempt. I then reduced the “matching thresholds” and it matched the lot. SongKong (& Jaikoz) did not match at all. Maybe something to do with matching thresholds?

I agree that 76% is OK, but a few more would be nice :wink: Using the MB id is a useful back-up, though.

BTW:
2,460 (76%) songs matched to MusicBrainz release
0 (0%) songs matched to MusicBrainz song only
430 (13%) songs matched to Acoustid song only
859 (26%) songs matched to Discogs
Only allow match if all songs in grouping match to one album:No
Only allow match if all tracks in album were matched:No

Thanks for the summary so you have already relaxed the musicbrainz criteria by unchecking Only allow match if all tracks in album were matched. But if you had full version and could save changes ran again over the same files I expect some of the 13% currently only matched to Acoustid song only would now be matched to MusicBrainz.

Ill look at the files when I get them, have not come through yet.

Basically there are two issues:

  1. The acoustids don’t link to any MusicBrainz releases so we cant make use of them.

  2. The artist, release and track metadata is all wrong so we failed to get a match using various combinations. To be more specific there was too much information so it seems the release metadata was

bach piano transcriptions vol 10 saint saens and philip

if it was just Bach Piano Transcriptions 10 we may have got a match,

We try various searches but there was similar problems with artist and track fields as well. Perhaps Picards matching is a little more lenient, I dont know its internal workings. But I’ll look to see if I could have tried something else to get a match.

You can tweak Picard to be as lenient as you like.

Jaikoz used to have such a feature many years ago but it didnt make much sense because as a user you don’t really know what you are changing when you make it more lenient and you could often end up allowing bad matches, so I removed it.

I prefer to just work to improving the algorithm so it works for different scenerios without breaking existing working senarios. This particular issue with very long release and track names containing people in titles is very much a Classical issue and something that I think I can improve on - https://jthink.atlassian.net/browse/SONGKONG-1168