Yes, but that's not what you are saying - you are saying that the test is meaningless for ordinary mortal's listening "these differences are insignificant to the point of meaningless."
Anyway, let's leave that be, OK?
Conclusive "Proof" that higher resolution audio sounds different
- Thread starter amirm
- Start date
Anyway, let's leave that be, OK?
Well, checking back in after another 20+ pages, all we read is a bunch of data free posts and solely anecdotal and opinion based argument about the ability of listeners to repeatably and reliably discern audible difference from long term listening. I guess that should come as no surprise ... that's what makes these forums go round and round.
Me, I'm still waiting for real data... and my Porsche.
Me, I'm still waiting for real data... and my Porsche.
What Tim is saying is exceedingly simple. So much so one must really go the long way round to see it differently.
Let us just imagine an audibility index. 100% is a difference large enough any person with normal hearing to 8 khz will hear it under nearly any conceivable condition where ambient noise doesn't exceed 100 db. 50 % on the index might be where 50% of everyone could hear it under not too terrible conditions. 1% might mean only 1% of listeners could hear it and only under highly favorable conditions of exceptionally low noise using SOTA playback equipment. Obviously glossing over many details and a highly contrived index. What Amir has heard might be 1%, might be .1% on such an index.
So even Amir will fail under noisier conditions, most will fail anyway, and it involved a highly unlikely signal scrutinized in a highly unlikely manner compared to simple music listening. Simply put we are getting into the very margins of where things will become completely un-hearable by human beings. (Assuming you do believe there are real limits to hearing acuity). So under almost all normal music listening coniditions this effect will not be heard even cannot be heard. Therefore, it is of no consequence an overwhelming percentage of conditions to an overwhelming percentage of people and one will not benefit any significant way to worry about it. I really find it hard to believe this isn't self evident unless one determines simply not to see it.
Let us just imagine an audibility index. 100% is a difference large enough any person with normal hearing to 8 khz will hear it under nearly any conceivable condition where ambient noise doesn't exceed 100 db. 50 % on the index might be where 50% of everyone could hear it under not too terrible conditions. 1% might mean only 1% of listeners could hear it and only under highly favorable conditions of exceptionally low noise using SOTA playback equipment. Obviously glossing over many details and a highly contrived index. What Amir has heard might be 1%, might be .1% on such an index.
So even Amir will fail under noisier conditions, most will fail anyway, and it involved a highly unlikely signal scrutinized in a highly unlikely manner compared to simple music listening. Simply put we are getting into the very margins of where things will become completely un-hearable by human beings. (Assuming you do believe there are real limits to hearing acuity). So under almost all normal music listening coniditions this effect will not be heard even cannot be heard. Therefore, it is of no consequence an overwhelming percentage of conditions to an overwhelming percentage of people and one will not benefit any significant way to worry about it. I really find it hard to believe this isn't self evident unless one determines simply not to see it.
We had the data regarding IM from JA (including using 30+khz tones), and of course the "data" (presentation) two of us linked by Dr David Griesinger.
TBH a lot of the thread has been taken up discussing IM and it has taken awhile to finally put that to bed; IM is not what Amir and a few of the others are picking out.
The only other data is the ABX passes; as we both chatted about near the beginning of this thread the ABX scope needs to be put into perspective; we know there is a difference now would be great to know why (Amir has a few thoughts and so do a few others including myself), until this is further investigated as we both mentioned and also a few others such as Amir nothing else can be taken from the test in regards how it may influence normal long term listening.
Cheers
Orb
We do have some data re the 2 tests put forth at AVS. Do we have any data regarding that which we first discussed 20+ pages ago? If so, I missed it.
TPDF is what used by studios/etc as this is what "should" be used (quotes as some still go their own way in pro world it seems), but you probably remember this is one area a few of us was thinking as a possibility.
Amir, you done further investigation into the dither aspect we all touched upon including you awhile back?
Cheers and thanks for the extra info esldude; would be amusing if TPDF is identified as being less transparent considering it is strongly recommended by various organisations/expert committees including such as at AES.
Orb
Amir, you done further investigation into the dither aspect we all touched upon including you awhile back?
Cheers and thanks for the extra info esldude; would be amusing if TPDF is identified as being less transparent considering it is strongly recommended by various organisations/expert committees including such as at AES.
Orb
Last edited:
Arny,
sorry but your case in response to JA on AVSF seems weak to me:
JA used two very compact USB stick DAC-Headphone amps that would suffer in performance due to size constraints, compounded by using USB and being also a DAC and headphone pre-amp all in one.
And yet they passed.
Sorry but if your gear is failing then it should never be used for any tests or even recording/measuring (appreciate you probably do not for this gear) as it would be unpredictable in terms of performance-behaviour when involving complex sounds in general.
You would slate any ABX test done with flawed equipment yes such as maybe DAC comparison one being NOS/two amps with radical spec differences with one pushed beyond spec/etc?
That said all those who currently past the test did so without IM issues (unless they pushed it into clipping/stress/strong distortion, and I wonder if you are doing same with your own gear as JA mentioned he had to reduce dbfs down).
Maybe I am jumping the gun here (not the first time for me and I would say nearly everyone in the audio hobby do at somepoint
), so knowing what the equipment was that failed would be great (I mentioned earlier the $30 Asus card has some issues as it is compromised to its higher model but I assume post #2501 on AVSF thread is possibly something else)
Thanks
Orb
sorry but your case in response to JA on AVSF seems weak to me:
ArnyK said:
JA used two very compact USB stick DAC-Headphone amps that would suffer in performance due to size constraints, compounded by using USB and being also a DAC and headphone pre-amp all in one.
And yet they passed.
Sorry but if your gear is failing then it should never be used for any tests or even recording/measuring (appreciate you probably do not for this gear) as it would be unpredictable in terms of performance-behaviour when involving complex sounds in general.
You would slate any ABX test done with flawed equipment yes such as maybe DAC comparison one being NOS/two amps with radical spec differences with one pushed beyond spec/etc?
That said all those who currently past the test did so without IM issues (unless they pushed it into clipping/stress/strong distortion, and I wonder if you are doing same with your own gear as JA mentioned he had to reduce dbfs down).
Stereoeditor said:
Maybe I am jumping the gun here (not the first time for me and I would say nearly everyone in the audio hobby do at somepoint
), so knowing what the equipment was that failed would be great (I mentioned earlier the $30 Asus card has some issues as it is compromised to its higher model but I assume post #2501 on AVSF thread is possibly something else)Thanks
Orb
Last edited:
It was my understanding, perhaps incorrect, that suggestions were to use TPDF in studios as it gives a fairly flat noise floor. The noise from the dither doesn't bunch up at any frequency. Then for the final stage or if you know you will only be doing one stage of dither that shaped was superior. Using shaped dither multiple times could cause the noise floor in the upper octave to become audible while it would remain quite low with TPDF.
Also, only of academic interest I suppose, I retrieved my inexpensive Asian condenser mics. Using an omni capsule, and making my own jangling keys recording at 192 khz, I was surprised how high the content went in frequency. Playing back the file at lower sample rates to make ultrasonics audible, I was surprised to find audible content between 35 and 40 khz. The content of jangling was quieter, but audible without any level boost. Even some content just above the noise floor between 40-50 khz with a 12 db level boost. The level between 25 and 30 khz was nearly as strong as that between 15-20khz. Of course I don't know the true noise profile of jangling keys. It may rise steeply with frequency.
ARC recommends 500 hour breakin on their tube stuff. Can you imagine having only 300 hours left before the sound goes down the crapper ? I can't.
You will find this dither article by Nika Aldrich really interesting (just be aware some technical question marks about noise shaping-etc but great if focus on dither applied and how looks - it is focused on studio engineering rather than EE DSP expertise): http://www.users.qwest.net/~volt42/cadenzarecording/DitherExplained.pdf
TPDF is always recommended as it is meant to be the optimum dither, with other considerations as well when dithering at/after mastering..
Also Keith Howard article (shame the more recent one is not online anywhere but was published in HiFiNews, latest one is really interesting and adds a bit more weight to his investigation IMO) touches on the various traits of PDF: http://www.stereophile.com/content/contingent-dither-page-3
BTW probably worth checking your jangling keys to both David Griesinger and James Boyk, they used either B&K 4133 or 4135 microphone, link to their own measurement for keys can be found in my post #233 this thread:
http://www.whatsbestforum.com/showt...ds-different&p=278622&viewfull=233#post278622
Big consideration I would say is input overload/microphone distortion/microphone correction, which may be more of a consideration for some other mics.
Just my take anyway.
Thanks again as it is great to read your own experience doing this as well.
Edit:
Just to add, bear Keith Howard's article relates to state of things back in 2005 (such as comment about quality of hirez releases-etc) when the article was published.
Thanks
Orb
Last edited:
Not quite. TPDF is an amplitude domain characteristic. It is orthogonal with spectral content or PSD.. What you seem to be talking about is TPDF, flat PSD dither. TPDF dither can have a flat PSD or not. There can also be non-TPDF dither, that has a flat PSD or not.
http://en.wikipedia.org/wiki/Power_spectral_density#Power_spectral_density
The good thing about TPDF dither is that it does a better job of randomizing the amplitude distribution of the the quantization error.
Generally speaking, flat PSD dither is suboptimal in terms of affecting the perceptual noise floor. With optimally shaped PSD dither, the subjective noise floor of a 16 bit recording can be close to 120 dB down as compared to about 93 dB down for dither with a flat PSD.
Here are the statistic for the keys jangling portion of the track:
The peak levels in the tracks are about -2 dB FS and -1 dB FS. This easily justifies using -0.5 dB FS signal as a test, including a small safety margin.
BTW here is the ABX log for me running an ABX test on just the keys jangling portion of the file:
--------------------------------
*Note - levels and passage selection fudged for best false positives
foo_abx 1.3.4 report
foobar2000 v1.3.2
2014/07/28 07:53:00
File A: C:\Users\client64\Music\AVS\Keys jangling\keys jangling full band 2496 test tones f3 4416.wav
File B: C:\Users\client64\Music\AVS\Keys jangling\keys jangling full band 2496 test tones f3.wav
07:53:00 : Test started.
07:54:38 : Trial reset.
07:56:40 : 01/01 50.0%
07:56:55 : 02/02 25.0%
07:57:15 : 03/03 12.5%
07:57:21 : 04/04 6.3%
07:57:27 : 05/05 3.1%
07:57:35 : 06/06 1.6%
07:57:42 : 06/07 6.3%
07:57:55 : 07/08 3.5%
07:58:10 : 08/09 2.0%
07:58:27 : 09/10 1.1%
07:58:35 : 10/11 0.6%
07:58:52 : 11/12 0.3%
07:59:09 : 12/13 0.2%
07:59:15 : 13/14 0.1%
07:59:22 : 14/15 0.0%
07:59:52 : 15/16 0.0%
07:59:59 : Test finished.
----------
Total: 15/16 (0.0%)
--------------------------------
Obviously, I'm not going to BS anybody, this was a bogus test. I selected the level and the portion of the track that I actually listened to to maximize the audible difference based on nonlinear distortion in the crappy monitoring system in this PC producing more audible IM with the 2496 test file than with the 4416 file.
And this was my reply to your post on AVS Arny .
Good morning Arny. Thank you for not calling me stupid again.
I am at a loss, yes I know that happens all the time, how that is the answer to the question I have asked. Those statistics do not tell you whether the peak amplitude is due to ultrasonic frequencies or not. This graph that I post readily does:
Clearly the test signals represent amplitudes way, way in excess of the same spectrum in the key jingling portion. So any distortion they cause would be far in excess of what the actual recording of keys did. As such it cannot be used as a diagnostic tool for why we could hear differences.
.Here are the statistic for the keys jangling portion of the track:
The peak levels in the tracks are about -2 dB FS and -1 dB FS. This easily justifies using -0.5 dB FS signal as a test, including a small safety margin.
Good morning Arny. Thank you for not calling me stupid again
. I am at a loss, yes I know that happens all the time
, how that is the answer to the question I have asked. Those statistics do not tell you whether the peak amplitude is due to ultrasonic frequencies or not. This graph that I post readily does:
Clearly the test signals represent amplitudes way, way in excess of the same spectrum in the key jingling portion. So any distortion they cause would be far in excess of what the actual recording of keys did. As such it cannot be used as a diagnostic tool for why we could hear differences.
Thanks for the info. It actually was the first link above from the cadenza recording PDF that I got the ideas on which dither to use. I have seen the same idea repeated elsewhere. On page 13 and 14 at the end of the article it says TPDF dither should be used at all stages with colored or shaped dither only being used in the final processing before "pressing a master". Been awhile since I read that so my thinking TPDF always gave a flat noise floor was a bit off.
Thanks for the links to jangling keys measured with good mics. Mine appear broadly similar until around 30 khz where my inexpensive mics roll off sooner and more steeply. Of course if mine were calibrated they could be used on signals somewhat higher in frequency. This is the part that surprised me as I expected them to start rolling off closer to their 20khz spec. That and the fact jangling keys are a simple way to check if something has some kind of response up to 40 khz or so.
I'll give the other articles a read later.
I am at a loss, yes I know that happens all the time
Clearly the test signals represent amplitudes way, way in excess of the same spectrum in the key jingling portion. So any distortion they cause would be far in excess of what the actual recording of keys did. As such it cannot be used as a diagnostic tool for why we could hear differences.
One very serious problem with your analysis is that you are apparently unknowingly comparing apples and tomatoes. Music is an incoherent signal and naturally shows up on a FFT as a large number of significantly lower data points.
Test signals are coherent signals and show up on a FFT as a small number of data points at far higher amplitude(s) when both signals contain the same amount of energy. Of course they don't have the same amount of energy in this situation for other reasons such as the vast difference in crest factors.
I suspect that my pure sine waves are actually too small to catch all monitoring systems that are causing audible distortion during these tests.
It appears that it is not unusual for headphone-based systems have fewer dynamic range reserves than loudspeaker-based systems.
Similar threads
