Tweak Musicbrainz Artist Query behaviour to improve accuracy of this fetcher

[ginjaninja 533]

Was looking through my library wondering why i had so many duplicate artists.

 

One of the main causes is some of my content is tagged by Picard (with musicbrainz ids embedded) and 2nd type is not tagged by Picard (albums which dont exist on musicbrainz)

so one gets correct match and second has to try pot luck and misses, and if the 2nd is scanned first, and the artist name is ambiguous, there is a strong possibility of a duplicate.

 

might it be possible to tweak the musicbrainz lookup so that when looking up an artist for the 1st time (with no embedded mbzid)

 

before querying the artist name alone

query the recording (track title) AND artist, to get artist id.

 

eg

https://musicbrainz.org/ws/2/recording?query=%22all%20of%20my%20heart%22%20AND%20artist:abc

 

and fall back to the "artist only" method in the unlikely event of no results.

 

would cut out a source of duplicate/mis-identifications i believe.

[Luke 37064]

@ginjaninja do you still have cases where you feel this would help?

[ginjaninja 533]

On 18/02/2021 at 02:47, Luke said:

@ginjaninja do you still have cases where you feel this would help?

i would not 'double down' on the assertion today that "querying musicbrainz by additional datapoints  will cut down duplicates", as i think the emby's scanning/fetching/matching algorithms may have changed since i last tested and made it so that artists with musicbrainz id are collapsed into artists without mbzid id.

but i would double down, that the accuracy of artist matches could be improved by using additional datapoints during musicbrainz querys (eg track title+artist)

i wrote a script as a proof of concept which i believe shows more accurate matching is technically possible. It had some good results even if limited to a fetch of available datapoints on the 1st track encountered (potentially artist+track title+album). ( this script uses its assessment on the best source metadata to search musicbrainz with, from across all tracks by artist in the database; not because that was required to improve accuracy particularly, but all things being equal , tracks on albums have a greater chance of being matched correctly with the additional album datapoint  -to search with)