Head-Fi.org › Forums › Equipment Forums › Sound Science › 24bit vs 16bit, the myth exploded!
New Posts  All Forums:Forum Nav:

24bit vs 16bit, the myth exploded! - Page 140

post #2086 of 2095
Quote:
Originally Posted by RRod View Post
 
Quote:
Originally Posted by castleofargh View Post
 

yup it was pretty interesting topic, even if you have to swim between the usual arny vs amirm fights.

from the little parts and indications from those who bought the paper, it's crap. they played the game of filtering at 16/44, used the only one dither method everybody said was bad for many years. had a very dynamic track obviously, with several moments with room sound and no music played. and it was with peaks around 100db or something.

 

so the trial was pretty much to know if we could identify 16/44 when everything was done wrong on a specific track. just the idea of filtering at 16/44 is such a joke. anyway, it's not the paper that will change the world, that much seems clear.

 

And I'm sure the quietest musical sound of the ensemble was well above the room noise. I mean, seriously, having played in them and listened to them live many times, string quartets aren't what should push 16 bits.


maybe they had big strong fingers? ^_^.

but sure I also don't expect the track to be extraordinary by itself. they say what it is somewhere but I can't recall.

post #2087 of 2095

I read the other forums and it's a lot of back and forth about whether it was peer reviewed, very little meat.  I think few (none?) of the posters had actually read the paper.

 

Why on earth are they filtering?  BTW, I already looked in-depth at the anti-alias filters used in modern 24 bit DACs last week.  The are rock solid.  Bottom line - with a 512 tap linear phase FIR which nowadays is nothing to implement in 180 or 65nm, the filters are rock solid.  Less than 0.1 dB passband ripple then falls off a cliff at 0.45 Fs.  No chance of any audible aliasing in the stop band (I think it was -130 dB or better).  

 

Anyhow, moving on.  One last try:

 

Some guy posted these files and a log showing 9/10 on Foobar ABX.  So I just did my regular SoX conversion 24 to 16 bit, still no dither (sorry Castle) and here we go again.  Still tantalizingly close.  Not as good as some other runs, since it shows some signs of mean reversion and longer runs of wrong guesses.  But still, after 40 trials, I'd at least like to be in double digits to be sure.  He had better phones so maybe that is the difference.  Sigh.  I would really like to see a conclusive test.  Maybe a selection of 5 top rated phones, 5 DACs covering all three 24-bit DAC makers and one or two computer DACs like Realtek, and maybe 100 listeners.  Best to record some content fresh to know the provenance.  It's a bit of work but not a major thing for pros, compared to all the dollars and effort put into marketing 24 bit audio.  

 

If anyone gets near 1% please post.


http://www.nordicaudiolabs.com/samples/24.wav
http://www.nordicaudiolabs.com/samples/16-hpt.wav

EDIT: Another 16 bit version made with Ditherbox (Airwindows Audio Unit Plugins) uploaded here for easy access:

http://www.nordicaudiolabs.com/samples/16-avd.wav

 

foo_abx 1.3.4 report
foobar2000 v1.3.6
2014/12/19 10:31:23

File A: C:\Users\Public\Music\HDtracks\Various Artists HDtracks Sampler\HDtracks 2014 Sampler\24.wav
File B: C:\Users\Public\Music\HDtracks\Various Artists HDtracks Sampler\HDtracks 2014 Sampler\24to16.wav

10:31:23 : Test started.
10:35:41 : 01/01  50.0%
10:36:32 : 02/02  25.0%
10:38:02 : 02/03  50.0%
10:38:43 : 02/04  68.8%
10:39:00 : 02/05  81.3%
10:42:01 : 03/06  65.6%
10:42:26 : 04/07  50.0%
10:42:58 : 04/08  63.7%
10:43:37 : 04/09  74.6%
10:45:42 : 05/10  62.3%
10:46:54 : 06/11  50.0%
10:47:28 : 06/12  61.3%
10:49:01 : 07/13  50.0%
10:49:22 : 07/14  60.5%
10:50:21 : 08/15  50.0%
10:50:39 : 09/16  40.2%
10:51:00 : 10/17  31.5%
10:51:24 : 11/18  24.0%
10:51:56 : 12/19  18.0%
10:52:53 : 12/20  25.2%
10:53:11 : 12/21  33.2%
10:54:12 : 12/22  41.6%
10:54:43 : 13/23  33.9%
10:55:08 : 14/24  27.1%
10:55:56 : 15/25  21.2%
10:56:07 : 15/26  27.9%
10:56:42 : 15/27  35.1%
10:57:04 : 16/28  28.6%
10:57:33 : 17/29  22.9%
10:57:57 : 18/30  18.1%
10:58:52 : 19/31  14.1%
10:59:24 : 20/32  10.8%
10:59:58 : 20/33  14.8%
11:00:46 : 20/34  19.6%
11:01:11 : 21/35  15.5%
11:01:32 : 22/36  12.1%
11:02:09 : 23/37  9.4%
11:03:34 : 24/38  7.2%
11:03:52 : 25/39  5.4%
11:04:26 : 25/40  7.7%
11:04:34 : Test finished.

 ---------- 
Total: 25/40 (7.7%)

post #2088 of 2095
Quote:
Originally Posted by castleofargh View Post
 


maybe they had big strong fingers? ^_^.

but sure I also don't expect the track to be extraordinary by itself. they say what it is somewhere but I can't recall.

 

Presto from Haydn Op.76 No.5 (one of the Erdödy quartets).

post #2089 of 2095
Warning: Spoiler! (Click to show)
Quote:
Originally Posted by Greenears View Post
 

I read the other forums and it's a lot of back and forth about whether it was peer reviewed, very little meat.  I think few (none?) of the posters had actually read the paper.

 

Why on earth are they filtering?  BTW, I already looked in-depth at the anti-alias filters used in modern 24 bit DACs last week.  The are rock solid.  Bottom line - with a 512 tap linear phase FIR which nowadays is nothing to implement in 180 or 65nm, the filters are rock solid.  Less than 0.1 dB passband ripple then falls off a cliff at 0.45 Fs.  No chance of any audible aliasing in the stop band (I think it was -130 dB or better).  

 

Anyhow, moving on.  One last try:

 

Some guy posted these files and a log showing 9/10 on Foobar ABX.  So I just did my regular SoX conversion 24 to 16 bit, still no dither (sorry Castle) and here we go again.  Still tantalizingly close.  Not as good as some other runs, since it shows some signs of mean reversion and longer runs of wrong guesses.  But still, after 40 trials, I'd at least like to be in double digits to be sure.  He had better phones so maybe that is the difference.  Sigh.  I would really like to see a conclusive test.  Maybe a selection of 5 top rated phones, 5 DACs covering all three 24-bit DAC makers and one or two computer DACs like Realtek, and maybe 100 listeners.  Best to record some content fresh to know the provenance.  It's a bit of work but not a major thing for pros, compared to all the dollars and effort put into marketing 24 bit audio.  

 

If anyone gets near 1% please post.


http://www.nordicaudiolabs.com/samples/24.wav
http://www.nordicaudiolabs.com/samples/16-hpt.wav

EDIT: Another 16 bit version made with Ditherbox (Airwindows Audio Unit Plugins) uploaded here for easy access:

http://www.nordicaudiolabs.com/samples/16-avd.wav

 

foo_abx 1.3.4 report
foobar2000 v1.3.6
2014/12/19 10:31:23

File A: C:\Users\Public\Music\HDtracks\Various Artists HDtracks Sampler\HDtracks 2014 Sampler\24.wav
File B: C:\Users\Public\Music\HDtracks\Various Artists HDtracks Sampler\HDtracks 2014 Sampler\24to16.wav

10:31:23 : Test started.
10:35:41 : 01/01  50.0%
10:36:32 : 02/02  25.0%
10:38:02 : 02/03  50.0%
10:38:43 : 02/04  68.8%
10:39:00 : 02/05  81.3%
10:42:01 : 03/06  65.6%
10:42:26 : 04/07  50.0%
10:42:58 : 04/08  63.7%
10:43:37 : 04/09  74.6%
10:45:42 : 05/10  62.3%
10:46:54 : 06/11  50.0%
10:47:28 : 06/12  61.3%
10:49:01 : 07/13  50.0%
10:49:22 : 07/14  60.5%
10:50:21 : 08/15  50.0%
10:50:39 : 09/16  40.2%
10:51:00 : 10/17  31.5%
10:51:24 : 11/18  24.0%
10:51:56 : 12/19  18.0%
10:52:53 : 12/20  25.2%
10:53:11 : 12/21  33.2%
10:54:12 : 12/22  41.6%
10:54:43 : 13/23  33.9%
10:55:08 : 14/24  27.1%
10:55:56 : 15/25  21.2%
10:56:07 : 15/26  27.9%
10:56:42 : 15/27  35.1%
10:57:04 : 16/28  28.6%
10:57:33 : 17/29  22.9%
10:57:57 : 18/30  18.1%
10:58:52 : 19/31  14.1%
10:59:24 : 20/32  10.8%
10:59:58 : 20/33  14.8%
11:00:46 : 20/34  19.6%
11:01:11 : 21/35  15.5%
11:01:32 : 22/36  12.1%
11:02:09 : 23/37  9.4%
11:03:34 : 24/38  7.2%
11:03:52 : 25/39  5.4%
11:04:26 : 25/40  7.7%
11:04:34 : Test finished.

 ---------- 
Total: 25/40 (7.7%)

 

 

You should have someone provide 2 tests for you, one with identical files and another with a 24bit and 24-16bit, just to see what your results might look like when the files are truly the same.  If you had no idea which test was which, it might help further support the notion that you're hearing a difference.

post #2090 of 2095
Quote:
Originally Posted by Greenears View Post
 

Where is the study/paper that proves that by experiment? It's so hard to find this stuff.


Have you seen this?

 

http://audioskeptic.blogspot.com/

 

And these?

http://www.aes.org/sections/pnw/ppt.htm

They should keep you occupied for a couple of weeks...

 

JJ Johnston arguably knows more than anyone else alive about the way the human auditory system works.

https://home.comcast.net/~retired_old_jj/

post #2091 of 2095
I do
Quote:
Originally Posted by Don Hills View Post


Have you seen this?

http://audioskeptic.blogspot.com/

And these?
http://www.aes.org/sections/pnw/ppt.htm
They should keep you occupied for a couple of weeks...

JJ Johnston arguably knows more than anyone else alive about the way the human auditory system works.
https://home.comcast.net/~retired_old_jj/
I don't see any test results in there. Did I miss something?
post #2092 of 2095
Quote:
Originally Posted by Don Hills View Post


Have you seen this?

http://audioskeptic.blogspot.com/

And these?
http://www.aes.org/sections/pnw/ppt.htm
They should keep you occupied for a couple of weeks...

JJ Johnston arguably knows more than anyone else alive about the way the human auditory system works.
https://home.comcast.net/~retired_old_jj/
huh? These are double blind tests in foobar. The method is well established. The only thing I want to see is that the tester creates the 16b file himself so we know the provenance. His post says he did.
post #2093 of 2095

Here's a little example I made for myself just now:

https://drive.google.com/file/d/0BwmVtb5IwniEVTlzWExvNW1QTlE/view?usp=sharing

 

The files are:

pluck.flac = 24bit

pluck16t.flac = 16bit truncated

pluck16d.flac = 16bit triangular dither

pluck16s.flac = 16bit shaped dither

pluck.sh = shell script that made the files

 

Basically I made a sequence of 1/2second-long synth plucks using sox, starting with 5s of them at full-scale (well more like -3dB) then doing a 1min long sequence of them fading to 0 at 24bits. I then resampled them 3 ways to 16bits. I did a few throw-away ABX trials and it wasn't looking good even for the truncated version, even if I cheated and pretended like the initial loud part didn't exist. But give it a try if you want; your results might vary.

 

I chose plucks b/c the make it easy to decide when you stop hearing the fade, which for me was at 41s for all of them if I set the pot to a decently loud level at the start (at 41s the levels are about -60dB peak/-72 rms). But if you then jack the dial again, you can start hearing the tone again (note: RRod takes no responsibility if you blow out your eardrums if you skip back to the beginning after you do this!).

 

Anyway, enjoy.

post #2094 of 2095
Quote:
Originally Posted by Greenears View Post

I do
I don't see any test results in there. Did I miss something?

http://www.aes.org/sections/pnw/ppt/jj/fund_of_hearing.ppt

Approximately 30 dB.

 

No test results, it was likely pre-war research.

I have access to the Bell Labs archives at work, I'll take a look...

post #2095 of 2095

Great read! Thank you

New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › 24bit vs 16bit, the myth exploded!