Well, the difference in reconstruction filter is more about the time domain than frequency domain performance (e.g. apodizing and linear phase, minimum phase and even more so in NOS dac etc). You don't need hearing to 20kHz to hear the (albeit subtle) differences in DAC reconstruction filters.
Have you tried actually performing the ABX test files from Goldensound? At least provide some DBT ABX test that you can distinguish with 90% correct answer (18/20) minimum between the two filters on his test file like I provided below:
Apparently, just did it for the very first time ever at least between two filters (but I've done lossy and lossless for the most part prior) and got 15/20 results. Slightly better than random guessing, but not sufficient enough that I can reliably tell although I do hear a very, very slight difference in spatial presentation and some extremely minor differences in the note edges (mostly ever so slightly noticeable in the 1-4KHz estimated range where one track is sharper and the other mellower) using the infamous distortion-laden Hifiman Susvara headphone
. But as you can see, I'm still imagining things since I didn't get 18/20 and p-value of 0.002. However, this is the closest to passing the single session DBT ABX test I've ever performed in my life.
Code:
foo_abx 2.2 report
foobar2000 v2.6.4
0124-05-09 21:32:38
File A: Test A (High Performance Filter).wav
SHA1: d626785e576b21b988a3ff3c59f85d3de27ed86d
File B: Test B (Normal Filter).wav
SHA1: 6cefd9bc846b7ba69d2bb06a869596cb740a4c0e
Output:
Core Audio : Yggdrasil+ [exclusive], 16-bit
Crossfading: NO
21:32:38 : Test started.
21:37:33 : Test restarted.
21:37:33 : 00/01
21:38:58 : Test restarted.
21:38:58 : 00/02
21:40:03 : Test restarted.
21:40:03 : 00/03
21:41:07 : Test restarted.
21:41:07 : 01/04
21:42:17 : Test restarted.
21:42:17 : 02/05
21:43:44 : Test restarted.
21:43:44 : 03/06
21:45:42 : Test restarted.
21:45:42 : 04/07
21:47:57 : Test restarted.
21:47:57 : 05/08
21:49:03 : Test restarted.
21:49:03 : 06/09
21:50:02 : Test restarted.
21:50:02 : 07/10
21:53:35 : Test restarted.
21:53:35 : 08/11
21:54:56 : Test restarted.
21:54:56 : 08/12
21:56:37 : Test restarted.
21:56:37 : 09/13
21:59:44 : Test restarted.
21:59:44 : 10/14
22:00:46 : Test restarted.
22:00:46 : 11/15
22:02:38 : Test restarted.
22:02:38 : 12/16
22:03:44 : Test restarted.
22:03:44 : 13/17
22:05:17 : Test restarted.
22:05:17 : 14/18
22:06:31 : Test restarted.
22:06:31 : 14/19
22:08:16 : Test restarted.
22:08:16 : 15/20
22:08:16 : Test finished.
----------
Total: 15/20
p-value: 0.0207 (2.07%)
-- signature --
c5e68ef4cc31b72cdd0fb36b601026280d83ec28
BTW, I could't find the abx log verification website again in case you guys want to prove that I didn't just made it up (I'm not that shameless), but I tried it on the abx page and it returned as valid
Referencing Amir's article,
https://www.audiosciencereview.com/forum/index.php?threads/statistics-of-abx-testing.170/
In my opinion, at least a 90% chance correct rather than 70% correct for 20 repetitions should be considered as credible of passing the ABX test