I agree with what he said, maybe less about the importance of the
reconstruction filter. In my experience that is very important for R-2R and essential for any Delta-Sigma DAC. I will talk only about the R-2R.
Now, that being said, he is right about the digital filters used for the OS. In my experience some of them are just bad, some are just OK... but a really great one has to be implemented by a dedicated DSP chip. A DSP can have a more refined OS algorithm (hopefully it is actually implemented).
A reconstruction filter for 44.1kHz that leaves the audio untouched is basically impossible to build because of the proximity of frequencies (20 VS 22.05kHz). Actually lots of the recordings are done at 96kHz to lessen the burden on that end (ADC).
The issue is because how a filter deals with the transients, pre and post ringing, intermodulation products and other effects (not very well understanded). Some of the filters have different characteristics (sharp roll off, slow roll off, etc), and each in the end is a compromise of some sorts between reconstruction (amplitude domain) and group delay (time domain). My guess is that some of us are more sensitive to the time errors (this is a brain function) and they would rather tolerate the imperfect reconstruction, that might be why some prefer NOS.
Actually I think that is also the reason why some (like me) like so much the sound of R-2R versus a D-S - the time domain issue that
plagues the D-S conceptually.
I think that an NOS player can still be used correctly (at higher sampling rates) using in front of it a dedicated OS solution (software on computer, DSP), followed by a minimal filter - that's possible on the higher frequency, it doesn't interfere with the audio band.
But you cannot go to the extreme of no filter. The gains on time domain are completely overtaken by the damage done in the analog domain. Some amplifiers simply cannot handle the extra ultrasonics pushed in them and will generate extra IMD. The tweeters might be overwhelmed too...
Now, for older people (40+), using good headphones (that can handle the ultrasonics), the ear cannot hear those high frequencies and it acts like a natural reconstruction filter at some 15-16 kHz, far enough from the 22kHz needed. I noticed that usually older people praise the NOS solution.
PS: About the Delta-Sigma (or Sigma-Delta, whatever school of thought you like) DAC's: Read the Sabre whitepaper that I have linked here (especially about the D-S modulator), you will see that the engineers are aware of the D-S issues and are trying to mitigate them while maintaining the production costs low. "Laser trimming" of an R-2R chip is not where they want go anymore... sadly.
Just a small passage, but there are more issues covered:
Another issue is this:
That made a lot of manufacturers (TI is one example) go towards the combined multi-bit ( for the higher bits) and D-S (lower bits) or made others to use multiple D-S blocks in parallel, thus effectively lowering the modulation on each of them.
However for a DSD (SACD) signal there is nothing you can do to "fix it". You can go higher in frequency (DSD128, 256, 512), pushing the garbage further under the rug.