Once you start to notice the many subtleties that go into a vocal performance, it is reasonable to ask “Is there a good way to describe them?”
We can think of this as a kind of arithmetic. Consider, for example, the “average” version of a song — some sort of mean of all the ways that different singers perform it. Then consider a particular artist’s rendition, and take the difference:
renditionthis artist – renditionaverage
What we’ve done here is remove information about the song itself, so that all that remains is whatever is unique to this artist’s performance.
We can go further, and compare a single performance by some artist against his/her aggregate performances:
renditionthis – renditionaggregate
Of course all of this assumes that we have some way of converting a song into a form in which such arithmetic works — or even makes sense.
Perhaps it would be interesting to then add that difference (in whatever representation you come up with) to the average rendition of a different song, to get a pseudo-performance by the artist. It sounds kind of like Image Analogies at that point.