Successful ABX testing to hear the difference between Redbook Audio vs upsampled to 192/24
Aug 16, 2013 at 6:54 PM Post #31 of 136

nick_charles

Headphoneus Supremus
Joined
Feb 26, 2008
Posts
3,180
Likes
334
Quote:
Wow, thanks for the analysis--what tools did you use?
 
Just Audacity and Excel but I am not convinced the differences I found would be big enough themselves to be audible, certainly although there was clipping it was only on a few samples, I was able to configure FooBar to allow me to ABX (ABX is just what FooBar calls DBT, DBT is short for Double Blind test, in a double blind test neither the experimenter nor the subject knows which stimulus is which, this stops the experimenter accidentally communicating the identity of samples, with FooBar the software takes the place of the experimenter.
 
Anyway I was unable to ABX the files in FooBar, however I don't know if the FooBar ABX comparator plays each file at its native bit-depth/sampling rate or whether it resamples. I set FooBar up to output a digital output to a DAC which supposedly can decode 24/192 - assuming it correctly output each one as it was encoded I failed miserably...
 
I don't think I have hearing over 20 kHz--when I run my own tests or use the various files online it disappears for sure around 18kHz at best.
No one has replicated my results (sorry, what's DBTS?), at least that I know of.
If you looked through the thread, I've passed the ABX under quite a few conditions now, and on two completely different platforms.

 
Aug 16, 2013 at 7:25 PM Post #32 of 136

xnor

Headphoneus Supremus
Joined
May 28, 2009
Posts
4,092
Likes
221
I just looked at the files. Yes, there is clipping in the 192/24 sample, but only a few samples.
 
The 44.1/16 file doesn't add up. When I resample it "properly" and subtract it from the 192/24 the remaining stuff is about ~50 dB down. When I subtract the downloaded 44.1/16 file the remaining stuff is only ~15 dB down.
 
Something went wrong during conversion. May be a simple alignment mismatch, but I don't have time to check that atm.
 
Aug 16, 2013 at 10:55 PM Post #33 of 136

UltMusicSnob

100+ Head-Fier
Joined
Aug 14, 2013
Posts
177
Likes
11
Quote:
I just looked at the files. Yes, there is clipping in the 192/24 sample, but only a few samples.
 
The 44.1/16 file doesn't add up. When I resample it "properly" and subtract it from the 192/24 the remaining stuff is about ~50 dB down. When I subtract the downloaded 44.1/16 file the remaining stuff is only ~15 dB down.
 
Something went wrong during conversion. May be a simple alignment mismatch, but I don't have time to check that atm.


What did you do to "resample it 'properly' "?
I checked the stats in SoundForge and you're right, they don't match, turns out in length. One is 41.889 secs, the other is 41.941. I clipped from files that had already been converted in their entirety, and didn't take care to make sure they were precisely the same length. It was just a selection error, not conversion.
 
Aug 16, 2013 at 10:59 PM Post #34 of 136

UltMusicSnob

100+ Head-Fier
Joined
Aug 14, 2013
Posts
177
Likes
11
Quote:
Just Audacity and Excel but I am not convinced the differences I found would be big enough themselves to be audible, certainly although there was clipping it was only on a few samples, I was able to configure FooBar to allow me to ABX (ABX is just what FooBar calls DBT, DBT is short for Double Blind test, in a double blind test neither the experimenter nor the subject knows which stimulus is which, this stops the experimenter accidentally communicating the identity of samples, with FooBar the software takes the place of the experimenter.
 
Anyway I was unable to ABX the files in FooBar, however I don't know if the FooBar ABX comparator plays each file at its native bit-depth/sampling rate or whether it resamples. I set FooBar up to output a digital output to a DAC which supposedly can decode 24/192 - assuming it correctly output each one as it was encoded I failed miserably...
 

 

Thanks for the info. I'm using the foobar WASAPI plugin, which is supposed to deliver "bit-perfect" digital streams to the DAC. 
 
I initially found training my ears to find a difference very difficult. It's *very* easy to go toward listening for tonal changes, which does not help. I get reliable results only when trying to visualize spatial detail and soundstage size, and I tend to get results in streaks. I get distracted by imaginary tonal differences, and have to get back on track by concentrating only on the perceived space and accuracy of the soundstage image.
 
Aug 17, 2013 at 6:13 AM Post #35 of 136

stv014

Headphoneus Supremus
Joined
Jul 17, 2011
Posts
3,493
Likes
269
Quote:
Originally Posted by xnor /img/forum/go_quote.gif
 
The 44.1/16 file doesn't add up. When I resample it "properly" and subtract it from the 192/24 the remaining stuff is about ~50 dB down. When I subtract the downloaded 44.1/16 file the remaining stuff is only ~15 dB down.
 
Something went wrong during conversion. May be a simple alignment mismatch, but I don't have time to check that atm.

 
I have already commented on this before, it is a simple delay by about -0.00000292 s (roughly half a sample at 192 kHz). With that corrected, the difference is much lower.
 
Aug 17, 2013 at 6:41 AM Post #36 of 136

stv014

Headphoneus Supremus
Joined
Jul 17, 2011
Posts
3,493
Likes
269
Try converting with this command (the link to the program is in my signature):
Code:
 resample.exe -r 192000 -k 0.000002923655 -fw 0 -ff 0.5 -fl -2400 Test_File_Foobar_Redbook.wav test_192.wav
Of course, the delay and the filter are not perfectly accurate, so there is still some difference, but it is generally within +/- 1 LSB for 16-bit resolution (except for the clipped samples, obviously).
 
Aug 17, 2013 at 7:18 AM Post #37 of 136

xnor

Headphoneus Supremus
Joined
May 28, 2009
Posts
4,092
Likes
221
Quote:
What did you do to "resample it 'properly' "?
I checked the stats in SoundForge and you're right, they don't match, turns out in length. One is 41.889 secs, the other is 41.941. I clipped from files that had already been converted in their entirety, and didn't take care to make sure they were precisely the same length. It was just a selection error, not conversion.

Try Adobe Audition, or SoX (can be found in the latest Audacity, foobar2000 resampler plugin, ffmpeg ... or standalone), one of iZotope's offerings, stv014's resampler, ...
 
If you want to resample entire files and cut them afterwards try to cut them so that the samples line up. Shouldn't be hard with 192 kHz and 44.1 kHz.
 
Aug 17, 2013 at 9:26 AM Post #38 of 136

stv014

Headphoneus Supremus
Joined
Jul 17, 2011
Posts
3,493
Likes
269
Try to ABX this file against Test_File_Foobar_192_24.wav. It is the 44.1 kHz version recorded from a sound card, and is level matched to the original (which did result in some clipping). So, it includes playback at 44.1 kHz, but for the ABX test both files are in 192/24 format. Since the resampled file is slightly longer, the files should not be compared at the end.
 
Aug 17, 2013 at 9:58 AM Post #39 of 136

stv014

Headphoneus Supremus
Joined
Jul 17, 2011
Posts
3,493
Likes
269
Quote:
How long is the delay? I'm running Win7, but I didn't have the WASAPI plugin (just now picked it up, thanks for the tip).
I've previously tried deliberately going for delay time as a cue, but no success--the way I was running it (foobar2000 plus ABX utility only) I couldn't find a delay--I've never had a 100% success.

 
Checking it again, it seems the audible difference might not actually be the delay, but rather there is a click when switching to a different sample rate. It is louder when switching to 44.1 kHz than it is when the new sample rate is 192 kHz, and the loud click before the playback apparently gave the illusion of a shorter delay.
 
Aug 17, 2013 at 10:29 AM Post #40 of 136

xnor

Headphoneus Supremus
Joined
May 28, 2009
Posts
4,092
Likes
221
Aug 17, 2013 at 11:01 AM Post #41 of 136

stv014

Headphoneus Supremus
Joined
Jul 17, 2011
Posts
3,493
Likes
269
Quote:
Why run it through an additional D/A/D process?

 
It is for testing 44.1 kHz playback through real hardware, without comparing files of different sample rate in foo_abx. It should not really matter if the signal is upsampled/reconstructed by software or hardware, but the OP claims it does.
 
Aug 17, 2013 at 2:13 PM Post #42 of 136

UltMusicSnob

100+ Head-Fier
Joined
Aug 14, 2013
Posts
177
Likes
11
Quote:
Try to ABX this file against Test_File_Foobar_192_24.wav. It is the 44.1 kHz version recorded from a sound card, and is level matched to the original (which did result in some clipping). So, it includes playback at 44.1 kHz, but for the ABX test both files are in 192/24 format. Since the resampled file is slightly longer, the files should not be compared at the end.


Thanks for the file, and especially for the processing support and tips. I will try this file today.
 
Not trying to be obtuse here, but I need help having the specific procedure and reasoning spelled out for me.
 
I have a file I just downloaded from the link labeled "Test_File_Foobar_Redbook_XonarD1.wav".
But it's not a Redbook audio spec file, it's 192/24.
 
As I understand the above, this file "is the 44.1 kHz version recorded from a sound card" - what was done here precisely?
Recorded from what, my original 44.1 kHz version?
Recorded how? From the analog out, sampled at 192/24?
 
So the reasoning is, capture an analog 44.1 kHz analog playback in 192, so that this playback can be directly compared to a converted 192?
 
From another post: "It should not really matter if the signal is upsampled/reconstructed by software or hardware, but the OP claims it does."
  I'm not clear on what is being claimed here, nor what is being referenced in the OP. Perhaps there's is something implied in the OP?
    On the latter, I would have thought I was at least implying that the software or hardware does NOT matter, not that it does matter--thus my tests on the cheap hardware.
 
To be clear--all I'm claiming is that I hear a difference between pairs of files, built using the process I describe, on the platforms I describe.
 
I don't generalize my results to ***anything***, as a matter of rigorous demonstration. Statistically I can never do that, I'm a sample n of 1.
I think the ability to distinguish is interesting and bears examination, especially given the range of test conditions, and the fact that there are MANY statements online that say this positive result is flatly impossible.
I have reasoned in several places that the superior difference I hear likely stems from the D/A processing that takes place at playback. So, a null result here would support that hypothesis?
   To further speculate, I'm wondering whether the filtering differences between 44.1 and 192 playback D/A processes result in temporal resolution differences in the resulting files.
 
Aug 17, 2013 at 3:06 PM Post #44 of 136

xnor

Headphoneus Supremus
Joined
May 28, 2009
Posts
4,092
Likes
221
My guess: 192/24 source file -> resample to 44.1/16 -> play through D/A -> record through A/D at 192/24 -> cut -> save.
 
 
As for hearing differences, well it really depends on what you hear if anything.
 
edit: I'm gonna read the thread now, cause it seems like I missed a lot..
 
edit2: Oh you didn't take a 192/24 file as source, but a 44.1/16 one and resampled that one to 192/24. Now I get it. Ignore my files. Of course stv014 also took the 44.1/16 files directly for playback.
 
Aug 17, 2013 at 3:06 PM Post #45 of 136

stv014

Headphoneus Supremus
Joined
Jul 17, 2011
Posts
3,493
Likes
269
Quote:
Originally Posted by UltMusicSnob /img/forum/go_quote.gif
 
But it's not a Redbook audio spec file, it's 192/24.
 
As I understand the above, this file "is the 44.1 kHz version recorded from a sound card" - what was done here precisely?
Recorded from what, my original 44.1 kHz version?
Recorded how? From the analog out, sampled at 192/24?

 
Your original 44.1 kHz/16-bit file was played without any processing on one sound card, the line output of which was then recorded by another sound card in 192 kHz/24-bit format. The recorded file was processed in software for the purposes of level matching and accurate synchronization (see also the first link in my signature). So, compared to your original test, it basically goes through an additional 192/24 A/D-D/A loop, while the 192/24 sample converted from the original will be played as before.
 
Quote:
Originally Posted by UltMusicSnob /img/forum/go_quote.gif
 
So the reasoning is, capture an analog 44.1 kHz analog playback in 192, so that this playback can be directly compared to a converted 192?

 
Yes.
 
Quote:
Originally Posted by UltMusicSnob /img/forum/go_quote.gif
 
To be clear--all I'm claiming is that I hear a difference between pairs of files, built using the process I describe, on the platforms I describe.

 
With more tests, there is a better chance to find out what may be causing the difference.
 

Users who are viewing this thread

Top