El Joven y El Mar from spanish album Revelación
YOUNG MAN & SEA from chinese album Revelation
For those don’t know the background, the 1st album is a translated album with translated lyrics of the chinese album, completely different version in Spanish from the 2nd.
These 2 recordings are obviously different, and I tried to submit them in picard, but they are always related to the same acoustID.
Why is that, how to fix it?
It’s been unlinked from Recording “El Joven y El Mar” by G.E.M. - MusicBrainz. I think it sometimes just doesn’t work automatically and Picard still finds them. It should resolve itself if it hasn’t already.
In discord, we had a discussion
<@outsidecontext>: You mean both track’s fingerprints get assigned the same acoust ID? So the two fingerprints on Track "8a056f2f-3643-4106-94c6-c74f8a1456cb" | AcoustID are for the Spanish and Chinese version respectively? If so, then this means the two songs are considered similar enough by AcoustID to be identified as the same. And that means you cannot use AcoustID to distingiush between them.
<@outsidecontext>: We could check whether the submission counts increas if you submit the spanish version again. And whether this would also relink the recording.
i submitted again, and the source increases from 8 to 9. Clearly there is a false positive.
And actually, this isn’t the only case in the same album, I just brought the first one.
Oh, if only someone with a young mind would volunteer for fingerprint repair at GSOC…
I’m just a dreamer.
There’s nothing to fix, AcoustID is working as intended. Collisions of subtle audio differences are by design, and there is no mechanism to manually split “similar” fingerprints.
Previously discussed in Same AcoustID causes Picard to recognize songs incorrectly, The concept of AcoustID, etc.
I understand your argument, but I also want to say there might be a small improvement needed.
From the fingerprint comparison
their differences is more than obvious to tell they should be treated as different recordings.
The more black there is, the more similar they are. I don’t develop AcoustID, I’m just explaining that is not an obvious difference.
what i try to tell is, their differences are not evenly distributed, but centered in middle, unlike the example you cited.
hope someday they will make it smarter.
You seem to be misunderstanding the purpose of AcoustID. It’s not meant to distinguish recordings but to identify files:
It seems what was said „unfortunately„ is the part we are talking about. And the misfortune is continuing.
It’s still useful for that purpose.
There are certain limitations. A stereo and a 5.1-mix are definitely different MB recordings but can’t be distinguished by acoustic fingerprints. But fingerprints of a live and a studio recording are easily distinguishable and that’s often very helpful (they will not end up in the same AcoustID).
I’ve even found manipulated recordings where a section was inserted to make the recording slightly longer. 
You experienced a lot in your post, interesting.