In development I have a new algorithm for matching to Musicbrainz which is ‘release based’, this considers all songs that appear to be from the same release because either their release metadata says they are on, or because they are in the same folder and then finds the match that matches all songs to one release. After this first pass it then trys to match remaining tracks using the less stringent release match that the release only matches some songs, if this fails it then falls back to the existing track by track matching.
I’m getting great results with this, but have a few questions that I think you guys might have an opinion on.
- Compilations.
The new algorithm finds the best match regardless of whether its a compilation. Now some of you really prefer to have the original release but the way the released Jaikoz deals with compilations by giving them a slightly lower score doesn’t really work because sometimes they outscore a non-compilation release and sometimes they don’t depending on the values of the other metadata.
There is also a different between various artist compilations, and single artist compilations and you might not want the first but be happy with the second. For example many artists release a series of singles and e.p.s at the start of their career, and their first album is just a compilation of these early releases, but many of their fans may not even be aware that this album is a ‘compilation’ and would be happy for this to match.
So I propose a ‘Do not Match to Various Artist Compilations’ and ‘Do not match to Single Artist Compilations’ options, if these are both enabled then Jaikoz will ignore any compilations completely. Or alternatively I could have ‘Prefer Do not Match to Various Artist Compilations’ and ‘Prefer Do not match to Single Artist Compilations’ option which would only match to compilations if could not find a match at all using the new algorithm and non compilation releases.
- Scoring
Currently we have a score out of 100 which represents how well your track matches to a Musicbrainz track.
In the new algorithm when set of tracks matches a release the scoring is split into a track-level score which details how well the release matches all the tracks, and a release score which details how well the release matches your preferences such as preferred country , earliest release date ecetera.
But if we cant we have to go back to track by track matching than the earlier scoring comes into play.
This presents a small problem for the current metadata matching options, because the score means something different for when we manage to match
to a whole release to when we just match some tracks to a release, or just single track to track. So we could have three different scores options OR we could just remove these options altogether and trust decisions made internally by Jaikoz, OR we could remove score options altogther and have ‘Only Match Full Release’ options for those of you who want to ensure only watertight matches.
Also, the score when matching by puid options are no longer necessary because matching by puid is a lesser option than matching by release.
If you have an opinion on this please reply as quick as possible because I want to get on with this as quick as possible 