AVS/AIX High-Resolution Audio Test: Ready, Set, Go! - Page 4 - AVS Forum
Forum Jump: 
 36Likes
Reply
 
Thread Tools
post #91 of 215 Old 07-02-2014, 05:25 PM
AVS Addicted Member
 
amirm's Avatar
 
Join Date: Jan 2002
Location: Washington State
Posts: 18,230
Mentioned: 3 Post(s)
Tagged: 0 Thread(s)
Quoted: 871 Post(s)
Liked: 518
Quote:
Originally Posted by Mfusick View Post
It doesn't. Almost no system does. Typical consumer home noise floor is probably 40db so you need another 93db on top of that.
The noise floor number is not proper. SPL meters are "perceptually blind." You can't take their one number and have it mean anything with respect to audibility. See this article I wrote on proper way to analyze room noise: http://www.madronadigital.com/Librar...amicRange.html

Quote:
Originally Posted by Mfusick
The only people that have this full 93db of range (without significant compression or distortion) is the guys running very high end audio gear and/or DIY or pro audio gear and also have soundproofed and treated rooms to go with it. Around AVS there is some but in real life it's a needle in a haystack for normal folks.
Well, normal folks would not be concerned with high resolution audio, nor be married to threads like this . That said, there is data in my article about this:



As you see the noise floor of average system surveyed in the sensitive part of our hearing is around 10 spl and best systems go well below that, below the threshold of hearing.

If you look to the left you see your "40 db" number but that only holds true for very low frequencies.

Amir
Retired Technology Insider
Founder, Madrona Digital
"Insist on Quality Engineering"
amirm is online now  
Sponsored Links
Advertisement
 
post #92 of 215 Old 07-02-2014, 05:37 PM
AVS Special Member
 
CharlesJ's Avatar
 
Join Date: Jan 2006
Posts: 3,332
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 144 Post(s)
Liked: 109
Quote:
Originally Posted by Scott Wilkinson View Post
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.
Maybe the coin was a bit off in balance?
CharlesJ is offline  
post #93 of 215 Old 07-02-2014, 05:42 PM
AVS Special Member
 
CharlesJ's Avatar
 
Join Date: Jan 2006
Posts: 3,332
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 144 Post(s)
Liked: 109
Quote:
Originally Posted by arnyk View Post
Hold that thought.



You can say that but it will be a snowy day in San Diego before I ever say such a thing.

....
You mean it never snowed there in your lifetime?

Maybe it is time. Or, GW will not allow it.
CharlesJ is offline  
post #94 of 215 Old 07-02-2014, 05:57 PM
isa
Member
 
Join Date: Jun 2014
Posts: 35
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 29 Post(s)
Liked: 17
Quote:
Originally Posted by Scott Wilkinson View Post
I think the underlying issue has been found: Most sample-rate converters add 0.1 dB of headroom during the conversion, which lowers the overall level of the file by that much, so putting one file through two SRCs lowers the level by 0.2 dB. So the question now is, should we boost the level of the track that will undergo the double SRC by 0.2 dB before or after downsampling to 16/44.1? I agree that adding it after re-upsampling will raise the noise floor slightly. It seems to me we should increase the level of the source track before downsampling, but I wonder what you think?

BTW, we will continue to use the SRC that Mark is familiar with—Sonic Studios Sonic Process—which is a state-of-the-art SRC.
As an SRC, Sonic Studios seems OK for this effort: its shortcomings in a sweep test compared to SoX should be inaudible for any home system playing 44.1k files per

Cliff notes version: black outside of the main sweep is good, reflections and colors other than black less good. The less-than-black background of Sonic HD seems well below any reasonable level of background noise. That it's not SOTA in this regard isn't a game-changer for this test. That it begins to turn red at the limits of a classical understanding of audibility limits is potentially an issue but is probably OK for 44.1k files. Barely. And not SOTA.

But none of that addresses the dropped samples and dB changes. Surely Sonic gives informed users the ability to not apply any dB changes. I recommend disabling or changing that setting in Sonic rather than applying a level change in another step. I have no explanation or solution for the dropped samples, but I hope Sonic has a setting to avoid that as well.

Edit: Scott: thanks for your patience and doggedness in dealing with all of this. No good deed goes unpunished
Attached Thumbnails
Click image for larger version

Name:	Capture.JPG
Views:	205
Size:	93.6 KB
ID:	146017  
andyc56 likes this.

Last edited by isa; 07-02-2014 at 06:02 PM.
isa is offline  
post #95 of 215 Old 07-02-2014, 06:06 PM
Advanced Member
 
andyc56's Avatar
 
Join Date: Feb 2013
Posts: 638
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 17 Post(s)
Liked: 205
Quote:
Originally Posted by isa View Post
But none of that addresses the dropped samples and dB changes. Surely Sonic gives informed users the ability to not apply any dB changes. I recommend disabling or changing that rather than applying a level change in another step.
I downloaded the manual (PDF here) and looked through it briefly. There doesn't appear to be any way to disable the 0.1 dB attenuation of each pass of SRC. What's the deal with the dropped samples? I missed that. Edit: I guess that's the time misalignment of the original and resampled?

Quote:
Originally Posted by Sonic Studio
The digital filter that Sonic Studio Process SRC uses exhibits a small
amount of amplitude ripple near the cutoff frequency (this is known as the
“Gibbs Phenomenon” and is a characteristic of all FIR low-pass filters). Left
uncompensated, this ripple could cause source material with high frequencies
recorded at high levels to become clipped.
For this reason, SRC reduces the gain at the input of the conversion by 0.1 dB
to avoid clipping. Repeatedly converting the same file therefore will gradually
reduce its overall level.
I thought it took a worst case of about 3 dB attenuation to prevent clipping of a full-scale signal from the Gibbs phenomenon? The quote above from the manual seems confused between time and frequency domains.

Last edited by andyc56; 07-02-2014 at 06:25 PM.
andyc56 is offline  
post #96 of 215 Old 07-02-2014, 06:29 PM
isa
Member
 
Join Date: Jun 2014
Posts: 35
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 29 Post(s)
Liked: 17
Quote:
Originally Posted by andyc56 View Post
What's the deal with the dropped samples? I missed that. Edit: I guess that's the time misalignment of the original and resampled?
Kees first identified the issue in post 25. I don't have access to the files right now, but I recall about an 80 sample deficit between the CDDA compared to the HR files (it varied a bit per pair of file pair, I think).
isa is offline  
post #97 of 215 Old 07-02-2014, 06:52 PM
isa
Member
 
Join Date: Jun 2014
Posts: 35
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 29 Post(s)
Liked: 17
Originally Posted by Sonic Studio
The digital filter that Sonic Studio Process SRC uses exhibits a small
amount of amplitude ripple near the cutoff frequency (this is known as the
“Gibbs Phenomenon” and is a characteristic of all FIR low-pass filters). Left
uncompensated, this ripple could cause source material with high frequencies
recorded at high levels to become clipped.
For this reason, SRC reduces the gain at the input of the conversion by 0.1 dB
to avoid clipping. Repeatedly converting the same file therefore will gradually
reduce its overall level.

Wow, that's a crude way to deal with the issue. SOTA resamplers use more sophisticated smoothing approaches to avoid the need for dB hacks. Like SoX, for example.

Edit: I screwed up capturing the quote from Andy while I'm typing from my phone. But hopefully you get the point.

Last edited by isa; 07-02-2014 at 06:57 PM.
isa is offline  
post #98 of 215 Old 07-02-2014, 07:02 PM
Advanced Member
 
andyc56's Avatar
 
Join Date: Feb 2013
Posts: 638
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 17 Post(s)
Liked: 205
The new forum software is buggy. If you do a quote as follows (replacing parentheses with square brackets):

(quote=Sonic Studio)Some text(/quote)

that will give the quote a white background. If plain quote tags are used without "=Some Name", the white background is not displayed, and one line break is removed after the quote too, making one's reply and the quote run together.
isa likes this.
andyc56 is offline  
post #99 of 215 Old 07-02-2014, 07:02 PM
Member
 
wgscott's Avatar
 
Join Date: Aug 2010
Posts: 91
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 8 Post(s)
Liked: 32
Quote:
Originally Posted by granroth View Post
So can people perceive ultrasonic frequencies? That's precisely the point of this test!
If so, then it is improperly formulated, because it won't be able to distinguish between a person perceiving an ultrasonic frequency and a person perceiving an aliasing artifact well within the audible frequency bandwidth. (Either of these in principle would be a difference between HD and redbook audio, but the latter is the only option that is reasonable, given our current understanding of the human physiology of hearing.)


Quote:
Originally Posted by granroth View Post
As far as I know, there are no reliable (mostly agreed upon) studies that show either one way or another whether it's possible. I'm very interested to see the results of all this.
There are a very large number that demonstrate that humans can't perceive ultrasonic frequencies (above 20kHz). Hence the term "ultrasonic." But you don't even need a silent dog whistle; your 22 kHz test tone will be more than adequate for a small child, and probably an 18kHz test tone would be adequate for most adults. (I can hear a 17 kHz mosquito if it gets right near my ear canal, and I am 51, have a lot of ear-wax, and a stereo most audiophiles would sneer at.)

Last edited by wgscott; 07-02-2014 at 07:07 PM.
wgscott is offline  
post #100 of 215 Old 07-02-2014, 07:30 PM
isa
Member
 
Join Date: Jun 2014
Posts: 35
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 29 Post(s)
Liked: 17
Quote:
Originally Posted by andyc56 View Post
The new forum software is buggy. If you do a quote as follows (replacing parentheses with square brackets):

(quote=Sonic Studio)Some text(/quote)

that will give the quote a white background. If plain quote tags are used without "=Some Name", the white background is not displayed, and one line break is removed after the quote too, making one's reply and the quote run together.
That's exactly what I screwed up. Not a bug I think, just my mistake. Thanks!
isa is offline  
post #101 of 215 Old 07-02-2014, 07:39 PM
isa
Member
 
Join Date: Jun 2014
Posts: 35
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 29 Post(s)
Liked: 17
Quote:
Originally Posted by wgscott View Post
If so, then it is improperly formulated, because it won't be able to distinguish between a person perceiving an ultrasonic frequency and a person perceiving an aliasing artifact well within the audible frequency bandwidth. (Either of these in principle would be a difference between HD and redbook audio, but the latter is the only option that is reasonable, given our current understanding of the human physiology of hearing.)
There are many real and potential issues with this test, but aliasing artifacts do not appear to be among them. If you still believe that, show your work on how you came to that conclusion with your detailed analysis of the sampling processes used. Don't bother with telling me what "aliasing" means. I get that.
isa is offline  
post #102 of 215 Old 07-02-2014, 07:43 PM
Member
 
wgscott's Avatar
 
Join Date: Aug 2010
Posts: 91
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 8 Post(s)
Liked: 32
Quote:
Originally Posted by isa View Post
There are many real and potential issues with this test, but aliasing artifacts do not appear to be among them. If you still believe that, show your work on how you came to that conclusion with your detailed analysis of the sampling processes used. Don't bother with telling me what "aliasing" means. I get that.
I am afraid you don't get it.
wgscott is offline  
post #103 of 215 Old 07-02-2014, 07:53 PM
isa
Member
 
Join Date: Jun 2014
Posts: 35
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 29 Post(s)
Liked: 17
Quote:
Originally Posted by wgscott View Post
I am afraid you don't get it.
OK, let's assume I don't get it. But aliasing issues are obviously fatal flaws to any (re)sampling effort, thus you've called out Mark and Scott as incompetent on this. Care to offer any evidence other than ambiguity and insults? And lest you forget, you're the one who made the claim of aliasing issues, so back it up.
isa is offline  
post #104 of 215 Old 07-02-2014, 08:12 PM
Member
 
wgscott's Avatar
 
Join Date: Aug 2010
Posts: 91
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 8 Post(s)
Liked: 32
I was doing nothing of the sort, as you obviously are aware. In the unlikely event that you are not a troll, I am simply suggesting that there are many people who believe there are audible differences between redbook and say 24 bit, 96kHz music who also accept that humans cannot hear above 22kHz (the cutoff for redbook). Two possible reasons are aliasing artifacts and Fourier truncation artifacts. There may be others that I am less aware of. Those artifacts could potentially show up if a 24 bit, 96kHz recording is sampled at 44.1 kHz (typically in the course of CD production, for example). It isn't an accusation of incompetence on anyone's part. It is simply the consequence of the Shannon sampling theorem (or Nyquist or whatever you want to call it).

The test would allow determining if anyone can tell the difference between redbook and higher-frequency recordings, which I believe is exactly what Scott said at the beginning. I am pointing out that there are (at least) two different explanations for that, in the unlikely event that people can detect a difference in a statistically significant way. Because of this, it is not a test of whether people can hear frequencies above 22kHz, because that is only one of (at least) two logical possibilities.
wgscott is offline  
post #105 of 215 Old 07-02-2014, 08:23 PM
Advanced Member
 
andyc56's Avatar
 
Join Date: Feb 2013
Posts: 638
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 17 Post(s)
Liked: 205
A properly implemented sample rate converter has a brick wall LPF at one-half the minimum of the initial and final sample rates. This eliminates aliasing. Software implementations can be extremely good at this, hardware implementations less good.
andyc56 is offline  
post #106 of 215 Old 07-02-2014, 09:02 PM
Member
 
wgscott's Avatar
 
Join Date: Aug 2010
Posts: 91
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 8 Post(s)
Liked: 32
Hypothetically, if the results were to come back positive, i.e., that people can reliably detect differences between redbook and HD audio, which of the following two explanations do you think is more likely:

(1) AVS has somehow managed to discover something new about human hearing that 200+ years of international audio physiology research has somehow missed

or

(2) There are possible minor imperfections, like rounding errors, in the numerical implementation of a brick-wall algorithm that might lead to small but nonetheless audible imperfections that might be detectable with standard audio equipment.


For the record, I think both are unlikely, but if forced to choose, I would say (2) is far more likely than (1).
Frank Derks likes this.
wgscott is offline  
post #107 of 215 Old 07-02-2014, 09:24 PM
Advanced Member
 
andyc56's Avatar
 
Join Date: Feb 2013
Posts: 638
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 17 Post(s)
Liked: 205
I think any positive outcome is pretty darned unlikely too, but if there were one, I'd suspect a different possibility - speaker (or headphone) IM distortion products.
andyc56 is offline  
post #108 of 215 Old 07-03-2014, 12:17 AM
AVS Addicted Member
 
arnyk's Avatar
 
Join Date: Oct 2002
Location: Grosse Pointe Woods, MI
Posts: 14,382
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 753 Post(s)
Liked: 1167
Quote:
Originally Posted by wgscott View Post
The proponents of high-resolution audio are not (in general) claiming you can detect (or need to be able to detect) sound higher in frequency than 22 kHz. Rather, they are suggesting that sampling at 44.1 kHz (truncating the signal at 22.05 kHz) produces aliasing artifacts (and perhaps Fourier truncation artifacts) within the (non-controversially) audible frequency range. Hence, if the proponents are right in this assertion, there is no requirement for specialized speakers or other equipment that reproduces ultrasonic frequencies.

The science behind that assertion is based on the presence of strongly non-linear processing inside the digital domain, which in general does not exist since nonlinearity in the digital domain requires a high degree of intentionality (i.e. some kind of contrived nonlinear element such as one that calculates some mathematical function of its input signal).

In general the digital domain is totally free of nonlinear distortion, linear distortion, and noise. Such examples of nonlinear or noisy elements that we actually find are always in the circuitry that interfaces with the analog domain, such as ADCs amd DACs.
arnyk is offline  
post #109 of 215 Old 07-03-2014, 12:19 AM
AVS Addicted Member
 
arnyk's Avatar
 
Join Date: Oct 2002
Location: Grosse Pointe Woods, MI
Posts: 14,382
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 753 Post(s)
Liked: 1167
Quote:
Originally Posted by andyc56 View Post
I think any positive outcome is pretty darned unlikely too, but if there were one, I'd suspect a different possibility - speaker (or headphone) IM distortion products.
That has turned out to be a real world concern that has affected previous experiments.
arnyk is offline  
post #110 of 215 Old 07-03-2014, 12:26 AM
Advanced Member
 
mightyhuhn's Avatar
 
Join Date: Nov 2013
Posts: 639
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 264 Post(s)
Liked: 90
Quote:
Originally Posted by Scott Wilkinson View Post
I think the underlying issue has been found: Most sample-rate converters add 0.1 dB of headroom during the conversion, which lowers the overall level of the file by that much, so putting one file through two SRCs lowers the level by 0.2 dB. So the question now is, should we boost the level of the track that will undergo the double SRC by 0.2 dB before or after downsampling to 16/44.1? I agree that adding it after re-upsampling will raise the noise floor slightly. It seems to me we should increase the level of the source track before downsampling, but I wonder what you think?

BTW, we will continue to use the SRC that Mark is familiar with—Sonic Studios Sonic Process—which is a state-of-the-art SRC.
downsample it to 24/44.1 then back to 24/96 correct the db level with the 24 bit. now dither it to 16/96 and change it back to 24 bit by adding a couple of zero.
mightyhuhn is offline  
post #111 of 215 Old 07-03-2014, 04:22 AM
AVS Addicted Member
 
arnyk's Avatar
 
Join Date: Oct 2002
Location: Grosse Pointe Woods, MI
Posts: 14,382
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 753 Post(s)
Liked: 1167
Quote:
Originally Posted by Scott Wilkinson View Post
I think the underlying issue has been found: Most sample-rate converters add 0.1 dB of headroom during the conversion, which lowers the overall level of the file by that much, so putting one file through two SRCs lowers the level by 0.2 dB. So the question now is, should we boost the level of the track that will undergo the double SRC by 0.2 dB before or after downsampling to 16/44.1? I agree that adding it after re-upsampling will raise the noise floor slightly. It seems to me we should increase the level of the source track before downsampling, but I wonder what you think?

BTW, we will continue to use the SRC that Mark is familiar with—Sonic Studios Sonic Process—which is a state-of-the-art SRC.
I was reviewing this thread and realized that was an intense discussion of differences that were pretty much within the /- 0.1 dB tolerance that ABX allows.
arnyk is offline  
post #112 of 215 Old 07-03-2014, 04:34 AM
Advanced Member
 
mightyhuhn's Avatar
 
Join Date: Nov 2013
Posts: 639
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 264 Post(s)
Liked: 90
even if it is fine people will cry about it so let them fix it. it's "like" the fact that the files are wave not flac. if the files are flac people will arg lossless compression is impossible you know the rest...
mightyhuhn is offline  
post #113 of 215 Old 07-03-2014, 04:36 AM
Member
 
Kees de Visser's Avatar
 
Join Date: Jan 2009
Posts: 36
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 20 Post(s)
Liked: 17
Quote:
Originally Posted by isa View Post
Wow, that's a crude way to deal with the issue. SOTA resamplers use more sophisticated smoothing approaches to avoid the need for dB hacks. Like SoX, for example.
Isa, AFAIK Sonic SRC uses fixed point calculations, which means that clipping a signal is destructive and final. Most modern SRCs use floating point, which allows lowering >0dBFS signals without loss when converting to fixed point at a later stage.
That said, I don't know if the modern Sonic Studio Process SRC is still fixed point. I've made the 3 testfiles for the infinitewave website on my old Sonic HD. They show that SRC was already very good some 15 years ago The output was 24 bit, which explains the tiny bit of noise in the black background, compared to some modern 64 bit competitors.
Sonic also made 3 different SRC versions on user requests. Some ppl prefer steep filters and accept the higher ringing that comes with it. Others prefer a more gentle filter and take the aliasing for granted, since the reasoning is that it ends up in the transition (garbage) frequency band close to Nyquist, which is expected to be not or hardly audible. The famous Weiss Saracon demonstrates this even better.
I personally think that most manufacturers simply provide their customers with the options, even if there is no scientific proof that either solution is better or worse.
Kees de Visser is offline  
post #114 of 215 Old 07-03-2014, 05:39 AM
AVS Special Member
 
Frank Derks's Avatar
 
Join Date: Jul 2002
Location: Region A,B,C
Posts: 1,894
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 101 Post(s)
Liked: 89
I checked the hires files for headroom to see if the 0.2dB gain can be applied before the resampling to 16 bit.


Just My Imagination
No clipping at all. more than enough headroom


Mosaic
Clips only once near the middle of the track
It's only a slight clip.
Enough headroom


On the Street Where You Live
Several (mild) clipping events in the last quarter of the track.
Imo a 0.2 dB gain would not make much difference. Clipping would still be mild after gain.


Any thoughts on this?
Frank Derks is offline  
post #115 of 215 Old 07-03-2014, 06:07 AM
Newbie
 
parkhopper's Avatar
 
Join Date: Jul 2009
Posts: 8
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 9 Post(s)
Liked: 10
Quote:
Originally Posted by wgscott View Post
Hypothetically, if the results were to come back positive, i.e., that people can reliably detect differences between redbook and HD audio, which of the following two explanations do you think is more likely:

(1) AVS has somehow managed to discover something new about human hearing that 200+ years of international audio physiology research has somehow missed

or

(2) There are possible minor imperfections, like rounding errors, in the numerical implementation of a brick-wall algorithm that might lead to small but nonetheless audible imperfections that might be detectable with standard audio equipment.


For the record, I think both are unlikely, but if forced to choose, I would say (2) is far more likely than (1).
This exactly why I think it was a mistake to limit the experiment to those only with amplifiers and speakers capable of producing ultrasonic frequencies. It assumes the cause of an effect that is still in question. Would it not be better to open it up and simply take that information into account?

Prior to Scott's original post I downloaded and compared the Mosaic files from Mark's FTP site, and with my NAD C316BEE integrated amplifier (only rated at 20-20k), heard subtle, but clear differences. Certain effects were such that, once I keyed in on them with several back and forth playings, they became easy to identify which was which in a single blind test.

I hope the test parameters are reexamined. I think it would help add to what we might learn.
parkhopper is offline  
post #116 of 215 Old 07-03-2014, 11:22 AM - Thread Starter
AVS Special Member
 
Scott Wilkinson's Avatar
 
Join Date: Sep 2001
Location: Burbank, CA
Posts: 1,292
Mentioned: 7 Post(s)
Tagged: 0 Thread(s)
Quoted: 303 Post(s)
Liked: 1081
Quote:
Originally Posted by Frank Derks View Post
I checked the hires files for headroom to see if the 0.2dB gain can be applied before the resampling to 16 bit.


Just My Imagination
No clipping at all. more than enough headroom


Mosaic
Clips only once near the middle of the track
It's only a slight clip.
Enough headroom


On the Street Where You Live
Several (mild) clipping events in the last quarter of the track.
Imo a 0.2 dB gain would not make much difference. Clipping would still be mild after gain.


Any thoughts on this?
I would not want to add to any clipping, so applying gain before SRC seems like not the best idea, at least in two out of the three tracks as you cite here. One question I have is, assuming that clipping is not an issue, does it make a difference if the gain is applied before or after the SRC? It seems to me that it wouldn't make a difference. If the gain is applied before the SRC, that would raise the noise floor, then the SRC would lower it. If the gain is applied after the SRC, the noise floor would first be lowered by the SRC and then raised by the gain. Either way, the result would be the same, wouldn't it? Or am I missing something?

Scott Wilkinson
AVS Editor
Scott Wilkinson is offline  
post #117 of 215 Old 07-03-2014, 11:26 AM - Thread Starter
AVS Special Member
 
Scott Wilkinson's Avatar
 
Join Date: Sep 2001
Location: Burbank, CA
Posts: 1,292
Mentioned: 7 Post(s)
Tagged: 0 Thread(s)
Quoted: 303 Post(s)
Liked: 1081
Quote:
Originally Posted by parkhopper View Post
This exactly why I think it was a mistake to limit the experiment to those only with amplifiers and speakers capable of producing ultrasonic frequencies. It assumes the cause of an effect that is still in question. Would it not be better to open it up and simply take that information into account?

Prior to Scott's original post I downloaded and compared the Mosaic files from Mark's FTP site, and with my NAD C316BEE integrated amplifier (only rated at 20-20k), heard subtle, but clear differences. Certain effects were such that, once I keyed in on them with several back and forth playings, they became easy to identify which was which in a single blind test.

I hope the test parameters are reexamined. I think it would help add to what we might learn.
You make an interesting point. Perhaps as long as people identify what's in their system, we should accept determinations from everyone and take the system performance into account.

Scott Wilkinson
AVS Editor
Scott Wilkinson is offline  
post #118 of 215 Old 07-03-2014, 11:41 AM
Member
 
dschlic1's Avatar
 
Join Date: Aug 2011
Location: Tampabay, FL
Posts: 110
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 4 Post(s)
Liked: 18
So have the new audio files been posted? I am planning to listen to the files this weekend. Perhaps a word about my test setup.


I have a Dell M6700 notebook computer which has an IDT High Definition Audio system installed on it. According to the configuration for the audio system it will go up to 24 bit 192000 Hz sampling:
Sound Format.png


My audio system is a Yamaha RX-V665 AV receiver with a Klipsch F-30 speaker system. I am going to connect the notebook computer to the AV receiver with either or both HDMI and analog signal cable. I hope this results in a valid test setup.
Attached Thumbnails
Click image for larger version

Name:	Sound Format.png
Views:	16
Size:	13.9 KB
ID:	146810  

Sony KDL-42V4100 LCD TV, Yamaha RX-V665 AV receiver, Sony PS3 slim, Klipsch F-30 speaker system
dschlic1 is online now  
post #119 of 215 Old 07-03-2014, 01:00 PM
AVS Special Member
 
Frank Derks's Avatar
 
Join Date: Jul 2002
Location: Region A,B,C
Posts: 1,894
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 101 Post(s)
Liked: 89
Quote:
Originally Posted by Scott Wilkinson View Post
I would not want to add to any clipping, so applying gain before SRC seems like not the best idea, at least in two out of the three tracks as you cite here. One question I have is, assuming that clipping is not an issue, does it make a difference if the gain is applied before or after the SRC? It seems to me that it wouldn't make a difference. If the gain is applied before the SRC, that would raise the noise floor, then the SRC would lower it. If the gain is applied after the SRC, the noise floor would first be lowered by the SRC and then raised by the gain. Either way, the result would be the same, wouldn't it? Or am I missing something?

Four options: (each processing step is between [ ] )


A [Gain +0.2(dither?) ] -> [-0.1 SRC16dither] -> [-0.1 SRC24 -0.1]


B [ -0.1 SRC16dither] -> [Gain +0.2] -> [-0.1 SRC24]


C [-0.1 SRC16dither] -> [0.1 SRC24] -> [Gain+0.2 (dither?)]


or D (4 steps)
[Gain +0.1] -> [-0.1 SRC16dither] -> [Gain +0.1 ] -> [-0.1 SRC24]


The SRC software applied the -0.1 gain change before the resampling and dither raises the 16bit noise floor slightly *after* the resampling.


Best practice is to apply processing in the 24 bit domain. This rules out option B
Touching the resulting 16 bit file and the 24 bit lo-res version with gain is not desirable. Purist may object to this.
The dither after the 24 to 16 conversion preserves some of the dynamic range of the 24bit file and applying gain may affect that.


Imo it's best to do the gain in the 24 bit domain on a copy off the hi-res file. It might raise the noise floor but, as it is hires, this noise floor should be lower than the resulting 16 bit file anyway.


Highly controversial but you could also consider to apply -0.2dB gain on the hires file first.
You can consider this as a mastering step to fit the purpose of the file.
Use that 'mastered' file in the comparison as the hires sample and create the lo-res version from it using any of the options listed above you like best.
This will negate the effect of the +0.2 gain could have on the clipped samples while preparing the lo-res version.




Best is to simply use SOX....
andyc56 likes this.

Last edited by Frank Derks; 07-03-2014 at 01:05 PM.
Frank Derks is offline  
post #120 of 215 Old 07-03-2014, 01:14 PM
AVS Special Member
 
mcnarus's Avatar
 
Join Date: Nov 2007
Posts: 6,166
Mentioned: 3 Post(s)
Tagged: 0 Thread(s)
Quoted: 119 Post(s)
Liked: 321
You don't do if five times in a row -- you transpose A and B once and then use that transposition during the test.

ABX tests don't work that way. A unit isn't A or B until you connect it, and then the software identifies it as A or B for you. The only way to screw it up would be to hit the A button when you meant to hit the B button. Making that mistake 5 times in a row would be evidence of mental deficiency. Even lack of motor control would be random.

Well, I guess we understand probability in different ways. I was always decent at math but I can't claim any high degrees specializing on that. If you have a mathematics degree, then I'd like to know what elementary error I'm making.

Yes, I've studied statistics—enough to interpret DBT results, at least—and the mistake you are making is to invent some preposterously unlikely occurrence, which you cannot assign any probability to, and insist that is it far more likely than something with a known, non-zero likelihood.

Put one more way -- if I only have a 3% chance of consistently guessing a series of answers and I do consistently get the answers, then it's a fool's bet to say that I guessed -- probability theory says that I am far more likely to have known the answers than just guessed them.

Yes, but that's if you get the answers correct. If you get all the answers incorrect, as in this case, that cannot possibly be the explanation.

In the case we're talking about, if you have some evidence of a flaw in the test set-up, then yes, that could be an explanation of a 0/5 result. But simply positing some such problem because you don't like the result of the test is intellectually dishonest.

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

mcnarus is offline  
Reply Audio theory, Setup and Chat

Tags
frontpage

User Tag List

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