Ok, let's try to get back to the topic which
@oivavoi originally posted:
Here is what he said:
"As for their dac testing, Svante describes their methodology in Swedish in this forum post:
https://www.faktiskt.io/phpBB3/viewtopic.php?f=9&t=66684&full=1#p1960394
What they do for dac testing is some elaborate scheme where they compare the output from the dac under test with the output from a super-duper dac, and they also do some resampling of the signals to be sure that any difference is due to the dac-part of the device, and not sampling differences or other things. (please don't rely on my sketchy summary here for details, it's a simplified version)
In their latest testing, it does seem that they are in fact able to detect several dacs using this test - meaning that they are not transparent in the absolute sense of the word. Still, differences are very slight, and it's questionable whether it would be noticeable when listening to music in normal mode. And they also find that they have indeed been unable to detect some very cheap gear in their DBTs (such as the Yamaha WXC-50).
My understanding is that this group is mainly a collection of audio enthusiasts of a perfectionist bent, some of who happen to be audio professionals (either in manufacturing or in research). They seem to be animated by a desire to find audio gear that is absolutely transparent - in much the same way as our very own
@amirm . All their activities are on a voluntary basis. But they don't seem to be animated by a desire to prove everybody else on the internets wrong, as strange as that may seem for some of us."
Here is Google translate of that post from his link:
"Well, almost. We let the test dac play a signal in 44.1kHz 16 bit. This is recorded on the best ADC we have found, which also gets the easy job to do it in 96 kHz. We also take the 44.1 / 16 signal and sampled it with an SRC that is far better than the reconstruction filter typically in DACar to 96 kHz. This gives two 96/24 files as level matched within the thousands of a dB, then we listen to these two files. So, actually, there is an A / B listening of A: test object + ADC against the B: reampler. And the fun of the test setup is that both the recorder and ADC have a much easier task than ADC and therefore you can guess that it is especially the ADC that is heard, if any. Earlier tests were done as a FE-listen of an AD-DA chain, and then AD and DA must go at the same sampling rate, which made it difficult to express specificly about the DAC at a detection. Furthermore, there was a risk that the delay as the AD-DA chain became detectable due to buffering. This was at least a real problem at least, and there was somewhere where the development of the new test method started. The method is described in Molt 4-2014, although we have since changed to a better recapitulation."
As can be seen in that post it has been explained how the tests were performed but no conclusions have really been given. In other words, I cannot really find correlation between what he posted and what has been said in that post from the forum.