Head-Fi.org › Forums › Equipment Forums › Sound Science › Successful ABX testing to hear the difference between Redbook Audio vs upsampled to 192/24
New Posts  All Forums:Forum Nav:

Successful ABX testing to hear the difference between Redbook Audio vs upsampled to 192/24

post #1 of 136
Thread Starter 

I've been doing ABX tests using the foobar2000 ABX utility.

 

Taking commercially produced CD's, I upsample them to 192 kHz @ 24 bits.

Then I ABX the original 44.1/16 file against the upsampled one.

The files are converted using Sound Forge 10's included tool from iZotope, their 64-bit SRC.

I've done it with the default settings (filter steepness 32, alias suppression 175) and with the "highest quality" settings on the slider (filter 150, alias 200), with equal success.

I've tested against pop music (MEG, album 'Room Girl', tracks 'G Ballad' and 'Groove Tube'), and against classical (Christopher Parkening performing Albeniz with London Symphony), with the same success.

Here's the truly weird part: I've been using my Beyerdynamic DT 770 Pro's through Schiit Asgard2 for the most part, but today I replicated my results again using cheapo earphone plugs, from the output jack on a plain-jane PC with no soundcard (just built-in motherboard chips)--same success. The plugs provide decent isolation, but that's about it in terms of quality.

 One sample result is listed below, I have several others.

http://i.imgur.com/UdWshKh.png

 

Does anyone else have a successful ABX result to cite for upsampling?

Does anyone have any suggestions for upsampler settings? Filter steeper or gentler? Anti-aliasing? etc

 

Comments, hearing test challenges welcome.

post #2 of 136

This is not too surprising. An audible difference could be caused by imperfections in the upsampling program, the computer's audio pipeline (which may process different bitrates differently), the DAC (ditto), or quantization errors with resampling (unlikely at 24-bit).

 

Given that upsampling algorithms cannot output more real info (rather than noise) than what it started with, if somehow it caused a subjective improvement, then it would be due to certain euphonic processing that happened as byproduct of the conversion. In that case we might as well try to identify that process and apply it directly.

post #3 of 136

Could you upload some 30 second samples of the original and converted sound ? Also, what happens if you convert the 192 kHz version to 44.1 kHz/24-bit, and then compare that to the original ?

post #4 of 136
Quote:
Originally Posted by JMS View Post

the computer's audio pipeline (which may process different bitrates differently)

 

Yes, sample rate conversion and other processing by the operating system should be avoided, if possible. It may also be worth checking if the files are level matched (maybe the converter reduced the volume slightly to avoid clipping, even if it is not very likely; lossy codecs tend to do this, however).

post #5 of 136
Thread Starter 
Quote:
Originally Posted by stv014 View Post

Could you upload some 30 second samples of the original and converted sound ? Also, what happens if you convert the 192 kHz version to 44.1 kHz/24-bit, and then compare that to the original ?

Thanks for the tips, I'll do the downsample test.


Edited by UltMusicSnob - 8/14/13 at 8:47am
post #6 of 136
Quote:
Originally Posted by UltMusicSnob View Post

 

The result at the link was obtained from this pair:

http://www9.zippyshare.com/v/42165780/file.html
http://www9.zippyshare.com/v/75650487/file.html

 

Do let me know when you get it, so I can Delete this post.

 

btw, when I click on 'Insert Image', I just get a brief popup ("please wait"), then it disappears and I still don't see how to get the image here. Any tips appreciated. 

 

I have downloaded both files already. To upload images, click on the image icon, then "Upload", select the file to be uploaded, wait for the upload to complete, and finally click "Submit" (or change the preview size first if you want to).

post #7 of 136

Your converted file appears to be OK, with matched level to the original, and good sample rate conversion quality. It seems to be delayed by about -0.00000292 s, but that should not be audible. Also, there are a few samples that are slightly clipped, but I do not think that is what is responsible for the positive ABX result either.

 

There might be some playback related issue, though, so check if there is also a difference if you convert the 192 kHz file back to 44.1 kHz (but still 24-bit). Make sure that the playback is bit perfect as well (not resampled by the OS). Another possible problem I can think of is different latency (delay before the playback starts) depending on the sample rate, if the difference is large enough, it could give a spurious clue (an online MP3 quality comparison had a similar issue before, where the high bitrate file was revealed simply by taking longer time to load). Or there may be some artifact (click etc.) that appears in the audio output only when switching sample rates.

 

By the way, what is the difference you hear in the ABX test ?


Edited by stv014 - 8/14/13 at 7:50am
post #8 of 136
Thread Starter 
Quote:
Originally Posted by stv014 View Post

 

Yes, sample rate conversion and other processing by the operating system should be avoided, if possible. It may also be worth checking if the files are level matched (maybe the converter reduced the volume slightly to avoid clipping, even if it is not very likely; lossy codecs tend to do this, however).


I ran Sound Forge 'statistics' tool on the original and converted files and found levels matched, out to 4 or 5 digits of precision. I don't know what my personal JND is on levels, but I'd be amazed and gratified if I could spot differences this small systematically, across multiple types of content, playback systems, etc. Even my db meter only gives me 3 digits.

 

 

I don't have professional equipment for blind tests. This is just me with foobar2000 ABX and my DAW. The iZotope SRC appears to have a good reputation, other tool suggestions are welcome.

 

The files are converted ahead of time in iZotope 64-bit SRC before opening foobar ABX.

post #9 of 136
Quote:
Originally Posted by UltMusicSnob View Post

 

I ran Sound Forge 'statistics' tool on the original and converted files and found levels matched, out to 4 or 5 digits of precision. I don't know what my personal JND is on levels, but I'd be amazed and gratified if I could spot differences this small systematically, across multiple types of content, playback systems, etc. Even my db meter only gives me 3 digits.

 

I have already acknowledged in a previous post that the files do indeed have matched levels, so that is not an issue.

 

Quote:

Originally Posted by UltMusicSnob View Post

This is just me with foobar2000 ABX and my DAW. The iZotope SRC appears to have a good reputation, other tool suggestions are welcome.

 

I have also confirmed it already that the SRC quality is fine. So, either you can hear the lowpass filter of the converter (it only starts to roll off at 20.5-21 kHz), or the difference is in the playback and is specific to your system. I would guess the second one is more likely, but more testing is needed to find out where the difference really is.


Edited by stv014 - 8/14/13 at 9:19am
post #10 of 136
Thread Starter 
Quote:
Originally Posted by stv014 View Post

Your converted file appears to be OK, with matched level to the original, and good sample rate conversion quality. It seems to be delayed by about -0.00000292 s, but that should not be audible. Also, there are a few samples that are slightly clipped, but I do not think that is what is responsible for the positive ABX result either.

 

There might be some playback related issue, though, so check if there is also a difference if you convert the 192 kHz file back to 44.1 kHz (but still 24-bit). Make sure that the playback is bit perfect as well (not resampled by the OS). Another possible problem I can think of is different latency (delay before the playback starts) depending on the sample rate, if the difference is large enough, it could give a spurious clue (an online MP3 quality comparison had a similar issue before, where the high bitrate file was revealed simply by taking longer time to load). Or there may be some artifact (click etc.) that appears in the audio output only when switching sample rates.

 

By the way, what is the difference you hear in the ABX test ?


Thank you for the response! I think my preferred decision procedure (below) coincidentally helps protect against cues due to time between playback when switching--I don't use continuous playback, so there's always  start/stop/restart, regardless, and my time to restart varies quite a bit. It's not easy to hear these differences, so I also occasionally break for a couple of minutes to let the synapses recover from over-familiarity with the incoming stimuli.

 

In the foobar2000 ABX interface I always use the option to set a 'Start' and 'End' position, with the player jumping back to the Start point every time, whether repeating a single version, or moving to an alternative (foobar provides A, B, X, Y--I tend to listen to a lot of repeats, generally). So this is not like a recording studio A/B where you might split a signal onto two paths and then seamlessly move back and forth between them while the song keeps playing. foobar2000 will apparently let you repeatedly click back and forth between A and B while the song just continues, but I don't like that method: it compares two different clips, essentially, since if you let the song just keep playing, you're comparing two different musical segments.

Instead, to truly compare precisely the same musical segment formatted two different ways, I UNcheck "Keep playback position when changing track". This way I'm comparing the same few seconds of audio against each other. I use this for both A/B and X/Y comparisons.

I've tried deliberately selecting my X or Y based on a perceived switching time, to *look* for the artifact, but I can't get any results at all that way.

 

The difference I hear is NOT tonal quality (I certainly don't claim to hear above 22 kHz). I would describe it as spatial depth, spatial precision, spatial detail. The higher resolution file seems to me to have a dimensional soundstage that is in *slightly* better focus. I have to actively concentrate on NOT looking for freq balance and tonal differences, as those will lead you astray every time. I actively try to visualize the entire soundstage and place every musical element in it. When I do that, I can get the difference. It's *very* easy to drift into mix engineer mode and start listening for timbres--this ruins the series every time. Half the battle is just concentrating on spatial perception ONLY.

post #11 of 136
Thread Starter 
Quote:
Originally Posted by stv014 View Post

 

I have also confirmed it already that the SRC quality is fine. So, either you can hear the lowpass filter of the converter (it only starts to roll off at 20.5-21 kHz), or the difference is in the playback and is specific to your system. I would guess the second one is more likely, but more testing is needed to find out where the difference really is.


Wow, I appreciate all the analysis and advice! I would actually like to get some good analysis tools, so if you don't mind I wonder what you're using, and what you think is the best. I nearly bought Ozone 5 Advanced last month, but decided I'd better learn more about what's out there. I expect to do some mastering work for a regional band (friend of mine) in the near future.

 

On "specific to your system", I have successfully distinguished X and Y on both a higher end (Beyerdynamic DT 770 Pro's through Schiit Asgard2 from RME Babyface) and completely different much lower end (cheapo earphone plugs, from the output jack on a plain-jane PC with no soundcard) signal chain. I've got lots of computers laying around (CIS prof), so there are more alternatives to look at; this would at best only establish that whatever I'm hearing is NOT unique to the home studio chain, but that does strike me as something that adds information to my understanding, especially if I maintain the ability across many systems. Any other ways to pursue this angle?

 

How could I test directly whether I'm hearing an effect of the lowpass filter of the converter? Is it possible to tweak something in the process and highlight that factor? Or get a better converter? I'm in iZotope 64-bit SRC right now.

post #12 of 136
Thread Starter 
Quote:
Originally Posted by JMS View Post

This is not too surprising. An audible difference could be caused by imperfections in the upsampling program, the computer's audio pipeline (which may process different bitrates differently), the DAC (ditto), or quantization errors with resampling (unlikely at 24-bit).

 

Given that upsampling algorithms cannot output more real info (rather than noise) than what it started with, if somehow it caused a subjective improvement, then it would be due to certain euphonic processing that happened as byproduct of the conversion. In that case we might as well try to identify that process and apply it directly.


So far my main surprise is that I'm having difficulty finding "partners in crime", meaning other posted results of direct 192/24 to Redbook Audio comparisons which are replicably positive (I have a bunch, under various conditions). One of the reasons I came to Head-Fi was that Googling "ABX" and audio turned up a *bunch* of possibilities in this site, so I hope to uncover some ABX positives, for *whatever* reason---if there are artifacts, I also want to get hold of those files, and listen to them.

 

Over in GearSlutz I've heard suggestions several times that I'm hearing some sort of effect along the lines of the ones mentioned here---but no one has yet posted a demonstration that they or someone they cite has actually demonstrated success in ABX, artifacts or no.

 

If there are artifacts in my files, then other people should be able to get them and hear them as well (for that matter, if a true quality difference is available in 192/24, then other people should **also** be able to hear that and pass ABX tests). If it's specific to my equipment, of course, I can't post my equipment for someone else to use, that would have to be onsite. But I am pursuing multiple platform tests, have passed two on systems of *extremely* different quality.

post #13 of 136
Quote:

Originally Posted by UltMusicSnob View Post

 

On "specific to your system", I have successfully distinguished X and Y on both a higher end (Beyerdynamic DT 770 Pro's through Schiit Asgard2 from RME Babyface) and completely different much lower end (cheapo earphone plugs, from the output jack on a plain-jane PC with no soundcard) signal chain.

 

Well, a "low end" system having audible problems that depend on the sample rate would not be that surprising. Also, are you using bit perfect output (e.g. the WASAPI plugin) in foobar2000, or the default, which I think may be DirectSound (possibly adding the Windows sample rate converter to the signal path) ?

 

It would be interesting to compare 192 kHz conversions that use different lowpass filters, because that is where an audible difference could theoretically exist, if it is not introduced (spuriously) by the playback system.


Edited by stv014 - 8/14/13 at 10:38am
post #14 of 136
Thread Starter 
Quote:
Originally Posted by stv014 View Post

 

Well, a "low end" system having audible problems that depend on the sample rate would not be that surprising.

 

It would be interesting to compare 192 kHz conversions that use different lowpass filters, because that is where an audible difference could theoretically exist, if it is not introduced (spuriously) by the playback system.


Those conversions would reside in the actual D/A converter circuits, yes? Can you describe what you would expect a lowpass filter effect to sound like, as a differential? I've been working with only spatial cues to get results, not tonal quality. I don't hear more "air", or "brilliance", or "shine", or anything systematically related to frequency balance at all--and when I try, my results go straight to random. "Depth" is not much of an engineering term, but that's what I hear; that, and "spatial precision".

 

Also, I'm stuck wondering why if it's possible to distinguish 192 and 44.1 on artifacts like *any* of the above, I can't find any posted examples pro OR con.


Edited by UltMusicSnob - 8/14/13 at 10:44am
post #15 of 136
Quote:
Originally Posted by UltMusicSnob View Post

 

Thank you for the response! I think my preferred decision procedure (below) coincidentally helps protect against cues due to time between playback when switching--I don't use continuous playback, so there's always  start/stop/restart, regardless, and my time to restart varies quite a bit.

 

I mean the time between clicking "Play", and when you actually hear the sound. I did a quick test now running foobar2000 with WINE (on Linux) using the WASAPI plugin in "push" mode, and indeed I hear a delay difference which I could easily ABX with 100% success if not starting the playback at the beginning of the files.


Edited by stv014 - 8/14/13 at 10:47am
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › Successful ABX testing to hear the difference between Redbook Audio vs upsampled to 192/24