ABX 24/96 vs 16/44.1

[GXAlan]

Decided to try ABX'ing 24/96 vs 16/44 copies of Skyfall. These are as purchased, so not the 24/96 downsampled.

Foobar -> Cambridge Audio DacMagic XS -> Lexicon LEP1 IEM with stock filter (which is an AKG K3003)

At first, I couldn’t tell the difference at all doing the practice A and B. The DacMagic XS has audible hiss so it’s not the most transparent DAC but finally I was able to find a passage where there was a difference. This is where I had the songs looping.

A - feel the Earth move and *then*
B - feel the Earth move and then

A had a bit more attack, seemed a bit more emphasized whereas B was smoother. More “natural” possibly although I have never heard Adele live/unamplified to know what her voice truly sounds like. Even then, it often required 4 or 5 passes back and forth between X and Y until I could be sure.

By about half way through the test the ABX’ing got easier and I started to wonder which track was the high res track, A or B. I was blinded to the file names and I felt that there was a paradox. A seemed to have more detail but B actually sounded more natural in the sense that I felt as A was emphasized compare to B as opposed to B being subdued compared to A. Again, “natural” being an estimate since I don’t know what Adele sounds like in real life.

After a while, it was easy to detect the difference. Basically would just need XYXY and I would get it. But then it got a bit harder to hear the difference.

At trial 14, my wife asked me something. I took off my earbuds and when I put them back on, I just couldn’t hear a difference anymore. I took off my earbuds and put them back in. I then could hear the differences again but it didn’t seem as prominent of a difference as before and finished the test. I am not sure if there was a subtle difference in how the earbuds were put in my ear or if I lost my ability to discriminate the differences once I was no longer focusing on those six words from the song.

Conclusions
1) Differences are incredibly slight. I would not have found either rendition to be poor nor would I be able to identify the ABX if I only had one guess during testing.

2) Given the ability to click on ABXY as needed, not rushing, and only clicking when a difference wasn’t heard, it is possible to get good scores. At one point the differences seemed to be night and day but this wasn’t maintained.

3) Note that I said difference. At first I thought A was 24/96 since the sharper sound must be higher resolution but as I listened more, I thought that maybe it was aliasing and it was actually 16/44. In the end, I would still characterize them as different not good vs. bad

4) In any case, speakers offer the biggest difference. Get the biggest ones you have space for . Those are clearly different sounding.

Next step will be to try my Topping D50s and Geshelli amp to see if that makes it easier to distinguish between the two.


————


foo_abx 2.0.6d report
foobar2000 v1.6.7
2021-09-12 12:35:56

File A: 01 - Skyfall.flac
SHA1: 1c9672b1703fb51669dedfae655328bb574bb113
File B: 01-Skyfall.flac
SHA1: 8b3bdd707c075330b85a396a2e526292df3d894e

Output:
Default : Headphones (2- CA DacMagicXS 1.0) [exclusive], 24-bit
Crossfading: NO

12:35:56 : Test started.
12:38:42 : 01/01
12:40:02 : 02/02
12:41:01 : 03/03
12:41:51 : 04/04
12:42:14 : 05/05
12:42:44 : 06/06
12:43:13 : 07/07
12:43:31 : 08/08
12:43:51 : 09/09
12:44:10 : 10/10
12:44:26 : 11/11
12:44:47 : 12/12
12:45:22 : 13/13
12:45:34 : 14/14
12:47:52 : 15/15
12:48:37 : 16/16
12:48:37 : Test finished.

----------
Total: 16/16
p-value: 0 (0%)

-- signature --
53d293236fdbf646a3dd2d0b61f9b4361602ef45

 

[Blumlein 88]

The one fly in this ointment is whether or not both were mastered the same. Please don't shoot me.

It would be easy to check out. Use Pkane's Deltawave which would let you compare those tracks even with different sample rates as it would do the resampling. You might find out the EQ is a little different on one version vs the other. Sometimes, much of the time, they master the hires version differently.

 

[Zensō]

This is great, thanks for sharing the results. I came to basically the same conclusion and settled on 16/44.1.

 

[GXAlan]

Tried using Deltawave but it looks like it has to downsample to do its comparisons. It really does look like the same master just one is filtered and one isn't.

I don't fully understand what a delta spectrogram is, but if you look at the Delta Spectrogram, you see that there are a few spikes early on, and the part of the song that I think I can hear is in the region of those 3 spikes.

If I ask it to export the difference, it's empty because the downsampled 96 kHz has no data.

So then I tried something new. Go as fast as I can, but can still listen to a, b and x. I get 14/16.

foo_abx 2.0.6d report
foobar2000 v1.6.7
2021-09-12 22:41:38

File A: 01 - Skyfall.flac
SHA1: 1c9672b1703fb51669dedfae655328bb574bb113
File B: 01-Skyfall.flac
SHA1: 8b3bdd707c075330b85a396a2e526292df3d894e

Output:
Default : Headphones (2- CA DacMagicXS 1.0) [exclusive], 24-bit
Crossfading: NO

22:41:38 : Test started.
22:44:49 : 01/01
22:45:04 : 02/02
22:45:38 : 03/03
22:45:45 : 04/04
22:46:13 : 05/05
22:46:20 : 06/06
22:46:26 : 07/07
22:46:33 : 08/08
22:46:39 : 09/09
22:46:45 : 09/10
22:46:51 : 10/11
22:46:56 : 11/12
22:47:03 : 12/13
22:47:12 : 13/14
22:47:27 : 13/15
22:47:32 : 14/16
22:47:32 : Test finished.

----------
Total: 14/16
p-value: 0.0021 (0.21%)

-- signature --
6fa822a98b84e6cc35f5fc22c293e8a59588d014


----------------------

So then I was really curious and tried one more test. This time, I only listed to X and I had to start from the beginning of the song and wait until the part I mentioned. Only get one chance per trial to listen to X and can never go back to A or B. This was tricky because I was trying to listen for T's being more prominent.

It was pretty fatiguing listening to the same track over and over, and I just couldn't convince myself that I could tell whether I thought the "T's" were a bit sharper or a bit duller at all. When I try to make my "best guess" with only one try, it's not as good as when I can go back and forth with A and B.

foo_abx 2.0.6d report
foobar2000 v1.6.7
2021-09-12 22:53:07

File A: 01 - Skyfall.flac
SHA1: 1c9672b1703fb51669dedfae655328bb574bb113
File B: 01-Skyfall.flac
SHA1: 8b3bdd707c075330b85a396a2e526292df3d894e

Output:
Default : Headphones (2- CA DacMagicXS 1.0) [exclusive], 24-bit
Crossfading: NO

22:53:07 : Test started.
22:55:36 : 00/01
22:56:27 : 01/02
22:57:18 : 02/03
22:58:09 : 02/04
22:59:00 : 02/05
22:59:51 : 03/06
23:00:41 : 03/07
23:01:25 : 04/08
23:02:08 : 05/09
23:02:58 : 06/10
23:03:49 : 06/11
23:04:38 : 06/12
23:05:30 : 06/13
23:06:22 : 07/14
23:07:12 : 08/15
23:08:01 : 09/16
23:08:01 : Test finished.

----------
Total: 9/16
p-value: 0.4018 (40.18%)

-- signature --
5eef015eb7ee21c50e31060a370b56e96ed5106e


Took a break and did it again. Same result.


foo_abx 2.0.6d report
foobar2000 v1.6.7
2021-09-12 23:31:41

File A: 01 - Skyfall.flac
SHA1: 1c9672b1703fb51669dedfae655328bb574bb113
File B: 01-Skyfall.flac
SHA1: 8b3bdd707c075330b85a396a2e526292df3d894e

Output:
Default : Headphones (2- CA DacMagicXS 1.0) [exclusive], 24-bit
Crossfading: NO

23:31:41 : Test started.
23:35:01 : 00/01
23:35:55 : 01/02
23:36:46 : 02/03
23:37:37 : 03/04
23:38:24 : 03/05
23:39:13 : 04/06
23:40:03 : 05/07
23:40:51 : 06/08
23:41:35 : 06/09
23:42:10 : 06/10
23:43:00 : 07/11
23:43:32 : 07/12
23:44:08 : 08/13
23:44:48 : 08/14
23:45:24 : 08/15
23:46:18 : 09/16
23:46:18 : Test finished.

----------
Total: 9/16
p-value: 0.4018 (40.18%)

-- signature --
f373415acf01a1175533607453271b7534a9b611

So the the conclusion I have made is that there is definitely listener fatigue. Over time, I feel as if I have lost the ability to discriminate the differences that I could earlier.

I have a headache from listening to the same thing over and over from a technical standpoint instead of enjoying the music, so I will try to take a break from it and then come back in a few days and do one more ABX test.

 

[samsa]

The delta spectrograph has frequency on the vertical axis and time on the horizonal axis. Red means the 16/44 is louder (in that frequency range, at that time). Violet means the 24/96 is louder. Green means they are equal. As you can see, the 24/96 recording is uniformly "louder" above 20kHz. But through most of the recording, at frequencies below 20kHz, the spectrograph is green. So they are equally loud at all audible frequencies. However, in the first minute or so of the song, that equality breaks down rather badly. Clearly, they are different; probably audibly so.

 

[voodooless]

Chopp off the first minute and try again.

 

[Zensō]

My takeaway from this is that if one has to try so hard to hear a difference, it just doesn’t matter. It’s good to go through the exercise though, to settle the question in one’s mind so they can get on with enjoying music.

 

[julian_hughes]

My takeaway from this is that if one has to try so hard to hear a difference, it just doesn’t matter. It’s good to go through the exercise though, to settle the question in one’s mind so they can get on with enjoying music.

Actually abx testing itself is very demanding, and not at all the same mental process we engage in when just listening to music in the normal way. The best lossy codec in my experience is aac (mp3, ogg vorbis and opus all fail on the same very well known killer samples such as eig) and I know I've identified differences between compressed (aac) and lossless by normal listening, easily and unprompted simply because I knew the pieces so well, but when I came to abx it was *hard*. Not impossible, and I did indeed identify the different codecs, but it was demanding, required great attention and concentration and was tiring. All this to verify to myself something I'd already noted perfectly casually! Anyway, I don't use lossy codecs for audio any more as storage and bandwdith are plentiful & cheap now. But I would absolutely hate to try to abx red book vs "hi res". I've also never noticed a difference so don't have the motivation to do so. Thank god.

 

[GXAlan]

My takeaway from this is that if one has to try so hard to hear a difference, it just doesn’t matter. It’s good to go through the exercise though, to settle the question in one’s mind so they can get on with enjoying music.

Agree, 100%. Those are my thoughts too. “Comparison is the thief of joy.“. I probably will never be able to listen to Skyfall without being critical or distracted by the sound of “Ts” instead of just enjoying the piece.

Speakers and room EQ make a bigger difference than anything, which really makes me wonder if I should sell everything and get a StormAudio or Trinnov setup.

 

[samsa]

Chopp off the first minute and try again.

Or, better, downsample the 24/96 file, and do the ABX test with that.

What you're doing here is testing whether you can tell the difference between two different masterings of the song, which is quite a bit different from whether you can tell the difference between 24/96 and 16/44.