What I hear vs simplistic specs ( entry DAC/Preamps)

This is just my SUBJECTIVE listening of four entry level combinations. What I hear for sure and what I may hear. Emphasis is on the "digital edge" specifically how it relates to vocal sibilance through loudspeakers.  No made up phantom air and imaging stuff the You-tubers have to make up when they can't describe the differences they hear or think they hear. These are all Delta Sigma ESS chips so the test is really the execution, not the technology.  I am using my own design MOSFET amp and my own bookshelf speakers, Seas/SB with a sub. FLAC from JRiver, -3dB WASAPI. My hearing is still pretty good, but not what it was when half my age. Obviously, there is some masking from the amplifier as I benched it in the .003% 1W @ 1 to 4K range when I built it.  It was still under .03% 60W into a resistive load. Pretty proud of that if I do blow my horn a bit. 

In order of spec, reported by ASR,  USB in to RCA out. Curious, inverse price.

Schiit Asgard/ESS    .002% THD+N    $300   ( if the Modi is a mark, the DAC itself is in the .0008% range)

JSD Atom+ stack     .00012%              $209

Topping DX3pro+    .00015%               $200

Sabaj   A10d            .00008%              $180

What do I hear?

Asgard is clearly harsher. Even without comparisons it is not as smooth. So much in fact, this is why I went looking for a better one.  Five years ago, this was actually a respectable performance at entry to mid level and it was a big step up from my old Wolfson.  Note, the JDS DAC into the Schiit sounded smoother than the internal, so the edginess is still DAC related. 

JDS+ a clear step up.  Not perfect, but enough to be questioning if it is electronic or source. The preamp is slightly the limiting factor. I wonder, with a one step cleaner amp is any difference to the Topping or Sabaj would be audible.  I should whip up a passive volume control and test.

Topping. Maybe a difference, maybe not. I doubt I could pass a 3 sigma blind test and doubt anyone could pass a legitimate 6 sigma against the JDS.  This unit is the one that changed my mind on CH-FI catching up.  Last SMSL and Toppings a few years ago went right back to Amazon.  I might have stopped here if the range of the remote did not tick me off. 

Sabaj.  I think it is smoother.  Night and day compared to the Schiit, but to the JDS or Topping?  Maybe. On carefully selected tracks, I bet I could do 80%, but that is not statistically significant. I was not actually expecting it to be as good as the Topping only hoping the remote worked a longer range. Topping has really stepped up recently.  It does, slightly.   As they are on sale cheaper than the Topping, a deal!

What does this tell me?  I think it reinforces specs can matter to a degree. The jump from .002 to .0002% seems significant.   Curious how our brain can extract differences when masking should cover everything up. In a vacuum, I suspect one could live with any above that level.     \

I did play with the filters on the Topping and Sabaj.  On headphones there were slight differences between the brick wall and the more moderate settings. Through the speakers I did not hear a difference. The Sabaj also has PPL settings for the input which I have not played with. It may matter more if poor input timing. 

Conclusion?.  Next is to compare the Sabaj to some reputable implemented alternative technology  DACs. Chord and R2R specifically @ 10 to 20 times the price.  Do the improvements in subjective sound come from actual attention from the companies, technology, or as a byproduct of SINAD chasing?  Don't know. Don't care but the result does track within this group.  Going to compare at a high end shop with boutique level amp and speakers. If I can't hear a clear difference there, I sure won't at home. If I can, guess they will sell me something. 

FWIW, changing the JRiver stetting to -3 dB was by far a bigger difference than the Schiit to the Sabaj!   Keep things in context. That suggests the digital filtering clipping on overshoot is a real thing.  A theory as it could be tested objectively. I suspect it has been tested and proven as Chord and RME mention them handling this internally.