The question first is whether the song is specific to an album or collection or whatever. If you have the ID for the song, can you trace it back to a specific origin? (I haven't looked at the diagrams that hard.) If so, then the next question is whether there is EVER a time where you would have two vocalists on a single identifiable song and NOT have either a duet or a named combo? And in the case of a named combo would you name the members individually anyway?
When I did my CD database some years back, I took the approach called entity-attribute-value (EAV) layout. My structure was more or less:
Recording (RecID, media format, album title, record label, serial number, year released, and a couple of other lesser details)
Selection (SelID, RecID as FK, OrdID as the position in the sequence of selections, title, duration in min:sec as a DATE field, a couple of other things) OrdID was 1, 2, 3, etc. representing 1st track, 2nd track, ...
Details (DetID, SelID as FK, Detcode i.e. what was represented, text value of same) This was the EAV table where <SelID> was the entity, <DetCode> was the attribute, and then the value was literal text.
DetailCodes( DetCode, name of thing being represented)
The detail codes were things like "Artist" "Group" "Composer" "Soloist" (used when "group" represented an orchestra) "Accompanist" "Cover Artist" (when I was feeling whimsical or the cover art was unusual), etc. The DetailCodes table included a "sort order" value that said put Artist, Group, or Orchestra first, Conductor second, Soloist third, etc. I had a hierarchy of presentation order.
To be honest, I didn't care that I had a duet, but I could have annotated that through the EAV table by having an attribute "DUET." That property would be present (with a blank value) if the song was a duet, but absent if it was a two-person group like "Hall & Oates." H&O billed themselves as a combo, not two independent singers in "duet" style. Case in counterpoint, Andy Bennett and K.D.Lang's album where they clearly were as a duet.
Note that this wasn't a "PURE" EAV because in a pure EAV, the title and duration that were on the "Selection" record would actually have been in the EAV table as two more attrbutes. But I made some placement decisions for report convenience.