Doesn't it mean that 6% hear it, since 50% would be the result of "guessing"?
No.
I gather from the discussion of the referenced paper (can't/won't fork over $ to AES for copy), that 160 trials were conducted.
Of those 160 trials, 90 positive discriminations were made. That is 0.56 or 56%.
It can be (and was!) shown that this result corresponds to a result that is non-random, with a confidence of 95%.
IOW, only 5 times out of 100 would these results be arrived at by chance alone: there is a
signal in the noise.
As
@Kal Rubinson said, this says little about the magnitude of the result.
If the results are according to the pic attached here (taken from another ASR thread), they are still unimpressive.
As I said before, if "hi-res" were all it's marketed to be, healthy hearing would be able to discriminate fully.
Every. Single. Trial.