-
Bug
-
Resolution: Duplicate
-
Normal
-
None
-
None
-
None
-
None
Not sure if I can post this here for future improvements or if I shouldn't, please let me know.
I got strange high-level info for the following:
Track is C.E.A.R.T.A from KNEECAP. Style is hip-hop/rap in gaelic.
https://musicbrainz.org/recording/65161f21-ae51-48f4-84c0-2f7f53662081
https://acousticbrainz.org/65161f21-ae51-48f4-84c0-2f7f53662081
Gender: Female @ 86.2% - 2 guys are rapping
GTZAN Model: Jazz @ 32.2% - quite strange
I don't yet understand all the magic in the background as I started getting into AB yesterday to submit results, if it's AI how I can "help" it by saying the gender is male for example or if this is a more statistical analysis then maybe reporting this will help?
- duplicates
-
AB-449 Problems with "gender" model / measurement
-
- Open
-
Thanks for reporting this, I'm going to close it in favour of AB-449, which covers some interesting points about gender that we should discuss. I'll add some comments there
Regarding the gtzan model specifically, we know that this doesn't give great results and mostly just says 'jazz'. Ignore this one for now. The way to help improve this is to build more representative datasets with which we can train models.