AVS/AIX High-Resolution Audio Test: Ready, Set, Go! - Page 2 - AVS Forum
Forum Jump: 
 36Likes
Reply
 
Thread Tools
post #31 of 221 Old 07-01-2014, 09:50 AM
AVS Addicted Member
 
arnyk's Avatar
 
Join Date: Oct 2002
Location: Grosse Pointe Woods, MI
Posts: 14,387
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 762 Post(s)
Liked: 1175
Quote:
Originally Posted by granroth View Post
I agree with Waldrep's interpretation of his results, in this case. To those who don't follow his blog/daily emails, he did the test with ABX Tester on a Mac and came up with a 0% success rate. He asserts that that is equivalent to getting 100% since he clearly just switched around A and B in his head.

That makes sense to me. If he truly couldn't tell the difference then we'd expect to get somewhere closer to 50% (but not exactly that, since ABX Tester doesn't work like that). One of the core ideas of ABX testing is that it's statistically unlikely to blindly guess which file is which with 100% accuracy. Well, it's statistically precisely just as unlikely to blindly guess them all incorrectly.

So rejecting his test because he got 0% on it is no different than rejecting it if he had gotten 100% -- both are identically unlikely to happen by chance.

(I'm not sure why Waldrep didn't just re-do the test a second time, just to alleviate this confusion)
Two wrongs ( scoring zero and misunderstanding the instructions ) don't make a right.

Bad logic is not a substitute for doing the right thing, which is redoing the test. If he obtained 100% right for real, then redoing the test should be a cakewalk.
arnyk is offline  
Sponsored Links
Advertisement
 
post #32 of 221 Old 07-01-2014, 10:00 AM
Senior Member
 
tomtastic's Avatar
 
Join Date: Sep 2012
Location: Wichita, KS
Posts: 428
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 78 Post(s)
Liked: 38
I did a basic test about a year ago, just because I was interested in how my hearing was holding up and what my range was. As you probably know, when you're younger, under 20, you have decent hearing if you have no impairments. As you get older, you naturally loose range. I wasn't sure if I could still hear above 18khz or not so I tried it. I checked my speakers range, which the specs indicated up to 24khz.

I had no problem hearing 18 or 20, at 21 and 22 it was harder to notice, but right in front of the speaker I could still hear it and it was quite painful as I recall. I won't be doing it again any time soon. I think 22 was my limit, that or I couldn't find any source material higher at the time.

Really, because of the pain, i don't see any real reason to hear above 20 khz, so if I loose a little more I'm not worried about it. I had my wife take my same test and she couldn't hear anything above 15 khz if I remember right. She has a slight hearing impairment and my dad had similar results. He's worked in noisy construction environments all his life. Not to mention, as you age you naturally loose range. I would like to be able to hear 15-18 khz and not loose that but I'd say above that, not really important to me.

The important thing is to protect your ears. If you use headphones, keep the volume low. If you work in noisy environments, wear hearing protection. Hearing damage usually doesn't occur all at once, it's gradual, and once it's gone, it's gone.

I'm not sure if it was mentioned here, I don't remember reading, but use caution when listening to content over 20khz, it gave me a splitting headache when I did it and I'm sure prolonged use would be damaging to your ears.
tomtastic is offline  
post #33 of 221 Old 07-01-2014, 10:10 AM
AVS Special Member
 
Randy Bessinger's Avatar
 
Join Date: Jun 2009
Posts: 1,338
Mentioned: 7 Post(s)
Tagged: 0 Thread(s)
Quoted: 63 Post(s)
Liked: 67
Quote:
Originally Posted by tomtastic View Post
I did a basic test about a year ago, just because I was interested in how my hearing was holding up and what my range was. As you probably know, when you're younger, under 20, you have decent hearing if you have no impairments. As you get older, you naturally loose range. I wasn't sure if I could still hear above 18khz or not so I tried it. I checked my speakers range, which the specs indicated up to 24khz.

I had no problem hearing 18 or 20, at 21 and 22 it was harder to notice, but right in front of the speaker I could still hear it and it was quite painful as I recall. I won't be doing it again any time soon. I think 22 was my limit, that or I couldn't find any source material higher at the time.

Really, because of the pain, i don't see any real reason to hear above 20 khz, so if I loose a little more I'm not worried about it. I had my wife take my same test and she couldn't hear anything above 15 khz if I remember right. She has a slight hearing impairment and my dad had similar results. He's worked in noisy construction environments all his life. Not to mention, as you age you naturally loose range. I would like to be able to hear 15-18 khz and not loose that but I'd say above that, not really important to me.

The important thing is to protect your ears. If you use headphones, keep the volume low. If you work in noisy environments, wear hearing protection. Hearing damage usually doesn't occur all at once, it's gradual, and once it's gone, it's gone.

I'm not sure if it was mentioned here, I don't remember reading, but use caution when listening to content over 20khz, it gave me a splitting headache when I did it and I'm sure prolonged use would be damaging to your ears.
Great post. Scott had a great podcast a couple weeks or so ago. Definitely, worth a listen. One thing I learned was that noise related hearing loss is CUMULATIVE so if you go to loud concerts, it could result in noise related hearing loss much later. Something to think about.


Sorry for OT post.
Randy Bessinger is online now  
post #34 of 221 Old 07-01-2014, 10:27 AM
AVS Special Member
 
RobertR's Avatar
 
Join Date: Dec 2000
Location: California
Posts: 6,235
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 227 Post(s)
Liked: 421
Quote:
Originally Posted by arnyk View Post
Two wrongs ( scoring zero and misunderstanding the instructions ) don't make a right.

Bad logic is not a substitute for doing the right thing, which is redoing the test. If he obtained 100% right for real, then redoing the test should be a cakewalk.
Perhaps it's a case of quitting while (he thinks) he's ahead.
RobertR is offline  
post #35 of 221 Old 07-01-2014, 11:33 AM
isa
Member
 
Join Date: Jun 2014
Posts: 37
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 33 Post(s)
Liked: 18
I found the same result as Kees de Visser for level differences:
Attached Thumbnails
Click image for larger version

Name:	Capture.JPG
Views:	354
Size:	43.7 KB
ID:	143714  
isa is offline  
post #36 of 221 Old 07-01-2014, 12:03 PM
Member
 
WayneJoy's Avatar
 
Join Date: Feb 2013
Posts: 137
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 47 Post(s)
Liked: 34
yes, level differences need to be 0 for a valid experiment. If not, the louder tracks will probably be picked.
WayneJoy is online now  
post #37 of 221 Old 07-01-2014, 03:55 PM - Thread Starter
AVS Special Member
 
Scott Wilkinson's Avatar
 
Join Date: Sep 2001
Location: Burbank, CA
Posts: 1,343
Mentioned: 7 Post(s)
Tagged: 0 Thread(s)
Quoted: 348 Post(s)
Liked: 1176
Quote:
Originally Posted by isa View Post
I found the same result as Kees de Visser for level differences:
Interestingly, the difference in levels is not consistent between the high-res and downsampled versions. (Of course, I won't say which is which!) I will work on replacing the files with level-matched versions.

Scott Wilkinson
AVS Editor
Scott Wilkinson is offline  
post #38 of 221 Old 07-01-2014, 04:14 PM
isa
Member
 
Join Date: Jun 2014
Posts: 37
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 33 Post(s)
Liked: 18
To prevent anyone from trying to infer which version is which for the revised files based on loudness info for the current files, I encourage Mark to use a (pseudo) random number generator to decide which of the revised A and B files are high res.

For example: http://www.random.org/
isa is offline  
post #39 of 221 Old 07-01-2014, 04:52 PM
AVS Special Member
 
CharlesJ's Avatar
 
Join Date: Jan 2006
Posts: 3,408
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 188 Post(s)
Liked: 117
Quote:
Originally Posted by Scott Wilkinson View Post
Interestingly, the difference in levels is not consistent between the high-res and downsampled versions. (Of course, I won't say which is which!) I will work on replacing the files with level-matched versions.
But, just wondering, two Bs are higher than the A. And, all 3 are almost exactly .2 dB different. Curious.
CharlesJ is offline  
post #40 of 221 Old 07-01-2014, 04:54 PM
AVS Special Member
 
RayDunzl's Avatar
 
Join Date: Sep 2013
Location: Tampa, Florida
Posts: 1,306
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 454 Post(s)
Liked: 556
Two suggestions for extra credit:

Make one file with seamless transistions between the two "resolutions" since they are both now 24/96.

Make one file with real 24/96 on the left and the mutated 16/44 on the right. Seamlessly reverse the allocation a few times.

You won't "know" when the change occurs while it plays unless you can hear it or have spent time figuring it out from the data.

I'll be back later...


1.5RQ > digits > OpenDRC-DI > DEQ2496 > DAC2 > KCT > FPB 350mcx > reQuest > Sweetspot
RayDunzl is online now  
post #41 of 221 Old 07-01-2014, 05:18 PM - Thread Starter
AVS Special Member
 
Scott Wilkinson's Avatar
 
Join Date: Sep 2001
Location: Burbank, CA
Posts: 1,343
Mentioned: 7 Post(s)
Tagged: 0 Thread(s)
Quoted: 348 Post(s)
Liked: 1176
Quote:
Originally Posted by isa View Post
To prevent anyone from trying to infer which version is which for the revised files based on loudness info for the current files, I encourage Mark to use a (pseudo) random number generator to decide which of the revised A and B files are high res.

For example: http://www.random.org/
I've used random.org many times, and I might again for the new assignments here. In fact, it's more than a "pseudo" RNG; it uses data from atmospheric noise to generate truly random numbers rather than an algorithm, which is pseudo-random.

For the current files, I physically tossed a coin to assign the files to A and B.

Scott Wilkinson
AVS Editor
Scott Wilkinson is offline  
post #42 of 221 Old 07-01-2014, 05:38 PM
Advanced Member
 
granroth's Avatar
 
Join Date: Jan 2012
Location: Gilbert, AZ
Posts: 620
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 178 Post(s)
Liked: 149
Quote:
Originally Posted by arnyk View Post
Two wrongs ( scoring zero and misunderstanding the instructions ) don't make a right.

Bad logic is not a substitute for doing the right thing, which is redoing the test. If he obtained 100% right for real, then redoing the test should be a cakewalk.
Indeed, he absolutely should have re-done it and yes, it throws a lot of doubt on the process when he doesn't understand how it works. HOWEVER, since math dictates that scoring 0% purely by chance is equally as unlikely as scoring 100% purely by chance, we can safely say that he did detect a consistent difference between the tracks. It doesn't matter for the purposes of this test if he preferred one or the other -- this is entirely about the ability to perceive any difference at all. Nothing about his (problematic) test suggests to me that he couldn't perceive such a difference consistently.

This topic is utterly fascinating to me because logic dictates that it should not be possible to perceive a difference. If people are perceiving (and note I'm not saying "hearing") a difference, then how and what is it that is happening?
granroth is online now  
post #43 of 221 Old 07-01-2014, 06:30 PM
AVS Addicted Member
 
arnyk's Avatar
 
Join Date: Oct 2002
Location: Grosse Pointe Woods, MI
Posts: 14,387
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 762 Post(s)
Liked: 1175
Quote:
Originally Posted by granroth View Post
Indeed, he absolutely should have re-done it and yes, it throws a lot of doubt on the process when he doesn't understand how it works.
Hold that thought.

Quote:
Originally Posted by granroth View Post
HOWEVER, since math dictates that scoring 0% purely by chance is equally as unlikely as scoring 100% purely by chance, we can safely say that he did detect a consistent difference between the tracks.
You can say that but it will be a snowy day in San Diego before I ever say such a thing.

Even if he aced the test in a good way, the next logical step would be to ask for a redo.
arnyk is offline  
post #44 of 221 Old 07-01-2014, 06:46 PM
Advanced Member
 
granroth's Avatar
 
Join Date: Jan 2012
Location: Gilbert, AZ
Posts: 620
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 178 Post(s)
Liked: 149
Quote:
Originally Posted by granroth View Post
HOWEVER, since math dictates that scoring 0% purely by chance is equally as unlikely as scoring 100% purely by chance, we can safely say that he did detect a consistent difference between the tracks.
Quote:
Originally Posted by arnyk View Post
You can say that but it will be a snowy day in San Diego before I ever say such a thing.
Arny, you are one of the leading opponents of high-res audio on this site and since you approach it from such a logical perspective, I deeply respect you for that. Too often people approach topics like this from a dogmatic "belief" point of view and it's always refreshing to see those who don't.

But I'll admit that you're puzzling me at the moment. Surely you are not saying that randomly guessing with two choices would differentiate between 0% and 100%? Since ABX Tester has five A/B tests, his likelihood of choosing all right is 3.125% -- his likelihood of choosing them all wrong is also 3.125%. Each is identically likely/unlikely.

Quote:
Originally Posted by arnyk View Post
Even if he aced the test in a good way, the next logical step would be to ask for a redo.
On that we completely agree.
granroth is online now  
post #45 of 221 Old 07-01-2014, 07:49 PM
AVS Special Member
 
mcnarus's Avatar
 
Join Date: Nov 2007
Posts: 6,183
Mentioned: 3 Post(s)
Tagged: 0 Thread(s)
Quoted: 123 Post(s)
Liked: 324
(quote)Surely you are not saying that randomly guessing with two choices would differentiate between 0% and 100%? Since ABX Tester has five A/B tests, his likelihood of choosing all right is 3.125% -- his likelihood of choosing them all wrong is also 3.125%. Each is identically likely/unlikely.(/quote)

If you're guessing, that's true--0% is as likely as 100%. But we don't know that he's guessing. The other possibility is that he heard something. But that's not really a possibility in the 0% case, unless he's a moron.

So that's the difference: In the 0% case, there's only one reasonable conclusion--really bad guessing. In the 100% case, there are two: good guessing, or he really heard something.

If you can't explain how it works, you can't say it doesn't.—The High-End Creed

mcnarus is offline  
post #46 of 221 Old 07-01-2014, 08:41 PM
Member
 
Kees de Visser's Avatar
 
Join Date: Jan 2009
Posts: 41
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 24 Post(s)
Liked: 18
Quote:
Originally Posted by Scott Wilkinson View Post
Interestingly, the difference in levels is not consistent between the high-res and downsampled versions.
Scott, I find the difference to be consistent. Can you please re-check ?
It would make sense though, since many SRCs create a bit of headroom to prevent filter ripple from creating overloads. It's possible that since there were 2 SRC processes (down and up), we're talking about two times 0.1 dB.

Last edited by Kees de Visser; 07-01-2014 at 08:49 PM.
Kees de Visser is online now  
post #47 of 221 Old 07-01-2014, 08:57 PM - Thread Starter
AVS Special Member
 
Scott Wilkinson's Avatar
 
Join Date: Sep 2001
Location: Burbank, CA
Posts: 1,343
Mentioned: 7 Post(s)
Tagged: 0 Thread(s)
Quoted: 348 Post(s)
Liked: 1176
Quote:
Originally Posted by isa View Post
I found the same result as Kees de Visser for level differences:
Is this a module of foobar2000? I'm a Mac guy, so I don't have much experience with foobar. In the Track Gain column, what is the dB measurement in relation to? It seems to be in relation to 0 dB, but what does 0 dB mean in this case?

Also, I assume that the dB value for each track is somehow measured over the entire duration of the track, but I don't know how. And how are the left and right channels taken into account?

Finally, what is the meaning of the numbers in the Track Peak column? Are they the percentage of full scale?

Thanks for helping me understand these results!

Scott Wilkinson
AVS Editor
Scott Wilkinson is offline  
post #48 of 221 Old 07-01-2014, 09:03 PM - Thread Starter
AVS Special Member
 
Scott Wilkinson's Avatar
 
Join Date: Sep 2001
Location: Burbank, CA
Posts: 1,343
Mentioned: 7 Post(s)
Tagged: 0 Thread(s)
Quoted: 348 Post(s)
Liked: 1176
Quote:
Originally Posted by Kees de Visser View Post
Scott, I find the difference to be consistent. Can you please re-check ?
It would make sense though, since many SRCs create a bit of headroom to prevent filter ripple from creating overloads. It's possible that since there were 2 SRC processes (down and up), we're talking about two times 0.1 dB.
According to the screen shot posted by isa, two of the double-SCR files show the increased level, and one high-res file shows the increased level. I have kept very careful track of which is which, but it's not impossible that I made a labeling mistake (though I had Mark double-check them before posting). Did you get the same results as isa?

Scott Wilkinson
AVS Editor

Last edited by Scott Wilkinson; 07-01-2014 at 09:24 PM.
Scott Wilkinson is offline  
post #49 of 221 Old 07-01-2014, 09:42 PM
Member
 
Kees de Visser's Avatar
 
Join Date: Jan 2009
Posts: 41
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 24 Post(s)
Liked: 18
Quote:
Originally Posted by Scott Wilkinson View Post
According to the screen shot posted by isa, two of the double-SCR files show the increased level, and one high-res file shows the increased level. I have kept very careful track of which is which, but it's not impossible that I made a labeling mistake (though I had Mark double-check them before posting). Did you get the same results as isa?
You should look at the track peak values (higher is louder). Sorry for being vague. I'm trying not to spoil the test for those who don't use an ABX application
Kees de Visser is online now  
post #50 of 221 Old 07-01-2014, 10:07 PM - Thread Starter
AVS Special Member
 
Scott Wilkinson's Avatar
 
Join Date: Sep 2001
Location: Burbank, CA
Posts: 1,343
Mentioned: 7 Post(s)
Tagged: 0 Thread(s)
Quoted: 348 Post(s)
Liked: 1176
Quote:
Originally Posted by Kees de Visser View Post
You should look at the track peak values (higher is louder). Sorry for being vague. I'm trying not to spoil the test for those who don't use an ABX application
I appreciate that, but I'm going to relabel and repost the files as soon as we figure this out. Am I correct that Track Peak is the highest momentary level, and that in the display posted by isa, it is expressed as a percentage of full scale? Also, how are the left and right channels accounted for in a single number? I'm surprised that Track Peak is the thing to look at rather than overall or average level.

What do you think of using integrated LUFS as the guideline for equal levels? In the normalized versions Mark just made, the maximum difference between the high-res and downsampled tracks is 0.05 LUFS. This unit of measure does not correlate exactly with dB; it's more akin to RMS power.

Scott Wilkinson
AVS Editor

Last edited by Scott Wilkinson; 07-01-2014 at 10:11 PM.
Scott Wilkinson is offline  
post #51 of 221 Old 07-02-2014, 12:14 AM
Advanced Member
 
granroth's Avatar
 
Join Date: Jan 2012
Location: Gilbert, AZ
Posts: 620
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 178 Post(s)
Liked: 149
Quote:
Originally Posted by mcnarus View Post
(quote)Surely you are not saying that randomly guessing with two choices would differentiate between 0% and 100%? Since ABX Tester has five A/B tests, his likelihood of choosing all right is 3.125% -- his likelihood of choosing them all wrong is also 3.125%. Each is identically likely/unlikely.(/quote)

If you're guessing, that's true--0% is as likely as 100%. But we don't know that he's guessing. The other possibility is that he heard something. But that's not really a possibility in the 0% case, unless he's a moron.

So that's the difference: In the 0% case, there's only one reasonable conclusion--really bad guessing. In the 100% case, there are two: good guessing, or he really heard something.
The 0% case can just as easily be explained by transposing the two versions in his mind and simply not re-checking. That is, say he listens to A and B and notes their differences. He then goes through the five randomized samples and does notice a difference each time... but consistently presses B when he meant A and vice versa because he transposed which version was which. That's not being a moron, that's just being mildly forgetful.

I maintain that neither 0% nor 100% can be easily explained by either good or bad guessing, since a 3% chance is too slim to use any kind of guessing as a pat answer. Possible, but unlikely. It's FAR more possible that he really did detect a difference of some sort consistently and just transposed his choices.

Who knows -- it's possible the difference he detected is precisely the level differences discussed in this very thread.
granroth is online now  
post #52 of 221 Old 07-02-2014, 07:50 AM
AVS Addicted Member
 
arnyk's Avatar
 
Join Date: Oct 2002
Location: Grosse Pointe Woods, MI
Posts: 14,387
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 762 Post(s)
Liked: 1175
Quote:
Originally Posted by Scott Wilkinson View Post
Is this a module of foobar2000? I'm a Mac guy, so I don't have much experience with foobar.
This is Replaygain which is a feature of a number of products including FOOBAR2000. Its purpose in life is to equalize the perceived loudness of tracks from different sources. While it produces correct information it may not be the best tool for the purpose at hand.

Quote:
Originally Posted by Scott Wilkinson View Post
In the Track Gain column, what is the dB measurement in relation to? It seems to be in relation to 0 dB, but what does 0 dB mean in this case?
Replaygain's track gain is the amount of gain that is required to cause the track to play at a predetermined uniform level. If a track is generally too quiet it is a larger number. If a track is about the right loudness to meet a predetermined standard, then it is a smaller number.

Quote:
Originally Posted by Scott Wilkinson View Post
Also, I assume that the dB value for each track is somehow measured over the entire duration of the track, but I don't know how.
Replaygain is pretty sophisticated so its measurement is done over the entire track and it is a relatively sophisticated estimate of perceived loudness.

Quote:
Originally Posted by Scott Wilkinson View Post
And how are the left and right channels taken into account?
Replaygain does not mess with channel balance for the obvious reason.

Quote:
Originally Posted by Scott Wilkinson View Post
Finally, what is the meaning of the numbers in the Track Peak column? Are they the percentage of full scale?
I think so. I don't know off hand if it is based on unweighted amplitude or whether it is a number that includes perceptual effects.

I did an independent downsampling of the "Street Where You Life" high res file and found that downsampling to 16/44 dropped its level by 0.2 dB. The resampler was CoolEdit Pro 2 which has a good reputation for accuracy. I feel comfortable attributing the difference to the loss of information > 22 KHz.

I am fairly comfortable saying that any statistically significant results obtained in a good clean DBT would probably not be due to the observed level difference.
arnyk is offline  
post #53 of 221 Old 07-02-2014, 08:10 AM
AVS Special Member
 
RichB's Avatar
 
Join Date: Nov 1999
Location: Massachusetts
Posts: 8,882
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 323 Post(s)
Liked: 269
Quote:
Originally Posted by Scott Wilkinson View Post
I appreciate that, but I'm going to relabel and repost the files as soon as we figure this out. Am I correct that Track Peak is the highest momentary level, and that in the display posted by isa, it is expressed as a percentage of full scale? Also, how are the left and right channels accounted for in a single number? I'm surprised that Track Peak is the thing to look at rather than overall or average level.

What do you think of using integrated LUFS as the guideline for equal levels? In the normalized versions Mark just made, the maximum difference between the high-res and downsampled tracks is 0.05 LUFS. This unit of measure does not correlate exactly with dB; it's more akin to RMS power.
The first post shows these samples with the same frequency response through the range covered by the format.
The HD version has more information at higher frequencies.
This may seem overly simplistic, but wouldn't it be expected that a program analyzing the signal would find a more signal in the HD track?
This seems to be the point of the test, if that information is inaudible, then the listener would not hear these sounds (even on systems capable of producing them ).

- Rich

Oppo Beta Group

Parasound A51 Black (for sale) | Oppo BDP-105D | Oppo HA-1 | Oppo PM-1 | Parasound A21 & A31 | Revel Salon, Voice, Studio | Velodyne HGS-15

Last edited by RichB; 07-02-2014 at 08:16 AM.
RichB is offline  
post #54 of 221 Old 07-02-2014, 08:24 AM
AVS Special Member
 
mcnarus's Avatar
 
Join Date: Nov 2007
Posts: 6,183
Mentioned: 3 Post(s)
Tagged: 0 Thread(s)
Quoted: 123 Post(s)
Liked: 324
The 0% case can just as easily be explained by transposing the two versions in his mind and simply not re-checking. That is, say he listens to A and B and notes their differences. He then goes through the five randomized samples and does notice a difference each time... but consistently presses B when he meant A and vice versa because he transposed which version was which. That's not being a moron, that's just being mildly forgetful.
If he did it once, yeah. But 5 times in a row? Come on.

I maintain that neither 0% nor 100% can be easily explained by either good or bad guessing, since a 3% chance is too slim to use any kind of guessing as a pat answer.
It's not a pat answer. It's a proper statistical inference. Black swans happen. Trying to rationalize them away is bad statistics. That's what you're guilty of here.

Possible, but unlikely. It's FAR more possible that he really did detect a difference of some sort consistently and just transposed his choices.
Really? What's the probability? We know the probability of bad guessing, and it's non-zero. You're just pulling a probability out of your nether reaches here.

If you can't explain how it works, you can't say it doesn't.—The High-End Creed

mcnarus is offline  
post #55 of 221 Old 07-02-2014, 08:31 AM
isa
Member
 
Join Date: Jun 2014
Posts: 37
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 33 Post(s)
Liked: 18
I agree with Kees' and Arny's comments. I'm not a replaygain or volume leveling expert, but with that stated, here's what I know that might help:

1) Replaygain as currently implemented in foobar2000 implements the EBU R128 and related specs, with one exception that I know of (which I'll note shortly). I know of nothing better for dealing with perceived loudness issues. I ran Replaygain just as a check to see what it said after seeing Kees' comment.

2) The track peak is a percentage of full scale for the higher peak on the left or right channel in these cases (don't know what it does for more than 2 channel). The peak value is just FYI - the peak value calculation shown does not affect the gain calculations. The peak info can be used for other processing to deal with potential clipping that can occur after applying positive gains.

3) The track gain is the volume adjustment performed by a player that understands and uses Replaygain track gain adjustments. It's an adjustment that was calculated to the reference volume of -18 LUFS. This is the only area that I'm aware of that deviates from the EBU R128 spec, which calls for a reference volume of -23 LUFS. This causes zero problems: under any circumstances you just need all the songs you want to level to have their gains calculated referencing the same reference level.

4) This is where it gets interesting. I don't know why there is about a 0.2 dB difference between the CDDA and HR files, and thus I don't know what to recommend. Since the Replaygain track gains for the HR files are consistently about 0.2 dB less than the CDDA track gains, this means that the HR files are about 0.2 dB louder than the CDDA files prior to any leveling, according to the perceived loudness specs of EBU R128. I also note that, to the best I can understand in reading the EBU R128 specs, there is no filter in the volume leveling processing to limit high frequency content in the calculations, so frequency content above 22k will be accommodated in the loudness calculations. So I see 2 possibilities, but there may be more:
a) As Kees noted, perhaps a 0.1 dB offset was introduced with each resampling or processing step
b) Since the energy spectrum below about 22k is (presumably) identical in each version of a song, and since the HR version has additional energy above about 22k, it seems possible that the HR song could always be perceived as louder by Replaygain or any other volume leveling processor that uses a perceived loundness calculation without a high frequency limit.

That the dB differences are about 0.2 dB makes me think that 4.a is the cause. That the dB difference is not exactly 0.2 dB in each case makes me think 4.b could be the cause. Or it could be a combo of the two. But I'm waaay out of my depth in even speculating what's going on.

Hope that helps.

Edit: perhaps TMI, but the formula I use to convert peak info to dB info is Axon's formula: http://www.hydrogenaud.io/forums/ind...st=0&p=715285&. There are many pieces to the EBU R128 spec, but the main document I read that gave me the impression there is no upper frequency limit to loudness calculations is https://tech.ebu.ch/docs/tech/tech3343.pdf

Last edited by isa; 07-02-2014 at 08:44 AM.
isa is offline  
post #56 of 221 Old 07-02-2014, 09:26 AM
AVS Special Member
 
Frank Derks's Avatar
 
Join Date: Jul 2002
Location: Region A,B,C
Posts: 1,897
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 103 Post(s)
Liked: 89
Quote:
Originally Posted by isa
...
So I see 2 possibilities, but there may be more:
a) As Kees noted, perhaps a 0.1 dB offset was introduced with each resampling or processing step
b) Since the energy spectrum below about 22k is (presumably) identical in each version of a song, and since the HR version has additional energy above about 22k, it seems possible that the HR song could always be perceived as louder by Replaygain or any other volume leveling processor that uses a perceived loundness calculation without a high frequency limit.

That the dB differences are about 0.2 dB makes me think that 4.a is the cause. That the dB difference is not exactly 0.2 dB in each case makes me think 4.b could be the cause. Or it could be a combo of the two. But I'm waaay out of my depth in even speculating what's going on.
....

In one sample the hires is louder but in another the lo-res is louder. This contradicts both 4a and 4b.


I suspect it's just rounding errors in the calculated track gain dB values.

Last edited by Frank Derks; 07-02-2014 at 09:53 AM.
Frank Derks is offline  
post #57 of 221 Old 07-02-2014, 09:33 AM
isa
Member
 
Join Date: Jun 2014
Posts: 37
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 33 Post(s)
Liked: 18
Frank, I've shown my work: it's time you show yours .
isa is offline  
post #58 of 221 Old 07-02-2014, 10:09 AM
AVS Special Member
 
Frank Derks's Avatar
 
Join Date: Jul 2002
Location: Region A,B,C
Posts: 1,897
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 103 Post(s)
Liked: 89
Quote:
Originally Posted by isa View Post
Frank, I've shown my work: it's time you show yours .

Already pm'd the work I done to Scott.


Identifying the files was extremely difficult. I concentrated at minute cues in the samples that gave me a clue about which was the hires sample.


I was not able to successfully ABX because at that moment I didn't know which was which and after the identifying session I wasn't able to concentrate on my effort.


After the pm to Scott I checked the file's with audition and I know that the 0.2 dB difference did not affect me in identifying the samples.
When I'm up to it I will do another attempt at ABX.
Frank Derks is offline  
post #59 of 221 Old 07-02-2014, 10:17 AM
isa
Member
 
Join Date: Jun 2014
Posts: 37
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 33 Post(s)
Liked: 18
I meant show your work with respect to your conclusion that one or more of the CDDA files measured louder than its HR version. I showed my work using foobar2000 and Replaygain that the HR versions all measured louder, but I have no idea which of the CDDA versions you found louder than the HR versions, and how you determined that.

Last edited by isa; 07-02-2014 at 10:23 AM.
isa is offline  
post #60 of 221 Old 07-02-2014, 10:21 AM - Thread Starter
AVS Special Member
 
Scott Wilkinson's Avatar
 
Join Date: Sep 2001
Location: Burbank, CA
Posts: 1,343
Mentioned: 7 Post(s)
Tagged: 0 Thread(s)
Quoted: 348 Post(s)
Liked: 1176
Quote:
Originally Posted by Frank Derks View Post
In one sample the hires is louder but in another the lo-res is louder. This contradicts both 4a and 4b.


I suspect it's just rounding errors in the calculated track gain dB values.
You are correct; the track gain differences are not consistent between high res and low res. That's what's so puzzling.

Scott Wilkinson
AVS Editor
Scott Wilkinson is offline  
Reply Audio Theory, Setup, and Chat

Tags
frontpage

Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off