I have the following worry about audio listening tests. Suppose the
meaningful variable is a property of an extended passage, not a short
snippet. Then a subject's failure to accurately distinguish or
re-identify may be due to an inability to retain the property in
memory.

I am thinking by way of contrast to visual examples. I just made two
prints of a photograph using different settings on my printer. I am
looking at the face of the subject and I can see that the contrast is
higher in one than in the other. That is a Gestalt property of a
meaningful chunk of the picture, not a property of a few pixels (cf.
notes). The difference with the musical case is that I can compare
the contrast of the two pictures directly, whereas in music no
immediate comparison is possible. At best I have to keep the property
in memory, and maybe the relevant variable is something not easily
retained.

Is the existing empirical confirmation for tests recommended in audio
based largely on visual data? If so, perhaps they rely on factors
that apply to the visual domain (i.e., possibility of immediate
comparison) but do not transfer easily to audio.

Or are there cases in the scientific literature in which the relevant
kinds of tests have been found valid to measure the detection of
Gestalt properties of aural, temporally extended signals?