Ten or twelve years ago, two students who were graduating from a Tonmeister course in Detmold, Germany did a carefully controlled A/B study of DSD vs. high-rate PCM using very high-quality converters, listening through electrostatic headphones as well as good loudspeakers. They found a "null result"--there wasn't statistically significant evidence that their listeners (who were professional musicians, recording engineers and professors in both fields) could distinguish audibly between the two on musical program material.
They wanted to submit their paper for publication in the AES Journal; I was referred to them, and acted as German-to-English translator. I found their work serious and convincing. Also, in my dealings with them I saw no hint of personal bias; it seemed to me that they had set out to investigate the question, rather than to justify any particular conclusion about it.
I don't know whether the paper is on line or not, but can try to find it. In any event, to me it called into serious question any claims for obvious superiority of DSD sound vs. PCM. If you can remember when DSD was first being introduced, people were claiming that it solved all the problems of digital audio--that it was "digital that sounded like the best analog" and so forth. I'm not sure where those people are or what they would say today, though.
--best regards
P.S.: I found the study in German:
http://old.hfm-detmold.de/eti/projekte/diplomarbeiten/2004/dsdpcm/ -- It is available in English from the AES as Convention Paper #6086 (
http://www.aes.org/e-lib/browse.cfm?elib=12799): "DVD-Audio versus SACD: Perceptual Discrimination of Digital Audio Coding Formats" by Blech, Dominic; Yang, Min-Chi. The summary is:
To study perceptual discrimination between two digital audio coding formats, "Direct Stream Digital" and high-resolution (24-bit, 176.4 kHz) PCM, subjective listening comparison tests were conducted with specially recorded sound stimuli in stereo and surround. To guarantee their reliability, validity and objectivity, the double-blind ABX tests followed three main principles: The signal chain should be based on identical audio components as far as possible; these components should be able to convey very high audio frequencies; and the test population should consist of various groups of subjects with different listening expectations and perspectives. The results showed that hardly any of the subjects could make a reproducible distinction between the two encoding systems. Hence it may be concluded that no significant differences are audible.