Misleading Measurements

[pkane2001]

2 minutes ago, The Computer Audiophile said:

 Paul says it has existed for quite a while. 

 

I'm losing track of what we are talking about. What data? Audibility thresholds? Of course this exists.

[The Computer Audiophile]

9 minutes ago, pkane2001 said:

 

I'm losing track of what we are talking about. What data? Audibility thresholds? Of course this exists.

Yes. Audibility thresholds. 

[pkane2001]

14 minutes ago, The Computer Audiophile said:

Yes. Audibility thresholds. 

 

This may be a good start, as a summary:

 

http://www.aes-media.org/sections/pnw/ppt/other/limitsofhearing.ppt

 

 

 

 

 

 

 

 

[kumakuma]

36 minutes ago, The Computer Audiophile said:

Yes. Audibility thresholds. 

 

This wouldn't work for the many folks on this site who can hear things below these thresholds or hear the effects of sound waves below these thresholds on sound waves above them.

[The Computer Audiophile]

5 minutes ago, kumakuma said:

 

This wouldn't work for the many folks on this site that can hear things below these thresholds or hear the effects of sound waves below these thresholds on sound waves above them.

Fortunately, this is in the objective sub forum and I'm talking about how objective people talk about items that are below the threshold. Surely objectivists can agree on a level of human hearing and below which nothing matters. 

[The Computer Audiophile]

7 minutes ago, pkane2001 said:

 

This may be a good start, as a summary:

 

http://www.aes-media.org/sections/pnw/ppt/other/limitsofhearing.ppt

 

That was a good start, but I can't put the data into the real world with respect to measurements. Sure I understand this sentence

 

 

and this one

 

 

 

 

But I can't figure out how the rest of it relates to items below 0 dB. My understanding is limited and I'm susceptible to being swayed by measurements that I don't understand, that may have nothing to do with what I hear :~)

[kumakuma]

1 minute ago, The Computer Audiophile said:

Surely objectivists can agree on a level of human hearing and below which nothing matters. 

 

Why do you believe this is possible? I've never seen any evidence suggesting that audiophiles can agree on anything.

[The Computer Audiophile]

2 minutes ago, kumakuma said:

 

Why do you believe this is possible? I've never seen any evidence suggesting that audiophiles can agree on anything.

Because I believe people are generally good and don't seek to solely disrupt discussions with endless ands, ifs, or buts. There are always edge cases and those don't really concern me for this discussion. If objectivists can agree that human adults can hear up to 20 kHz, they can surely agree on other limits when presented real data. 

[Josh Mound]

9 minutes ago, pkane2001 said:

This may be a good start, as a summary:

 

http://www.aes-media.org/sections/pnw/ppt/other/limitsofhearing.ppt

 

 

 

 

 

 

 

 

I followed the ASR attempt to establish agreed upon thresholds for a while, and it seems like it petered out. My sense of why that happened is that we can mean “audibility threshold” in two different senses: what humans physiologically can detect and what research has established listeners hear in audio reproduction. For example, the audibility threshold of distortion was stricter in the initial ASR threshold attempt than in Archimago’s recent experiment. So, over at ASR there seemed to be a mix of physiology, AES-type research, and random assertions used to support the “lenient” and “strict” thresholds. 
 

Some of the results, such as a “strict” linearity of 120 seemed far beyond realistic audibility IMHO. At the same time, I kind of think IMD is more audible than that ASR post suggested. However, it definitely seems like there’s a lack of good research on some topics.
 

Then, of course, we’d get into issues of age and hearing and trainer versus untrained listeners.

 

Ultimately, this uncertainty and complexity is why I think a lot of people default to the “moar is better” approach. But I think a “these DACs all clear the bar of flat FR, linearity to 90 dBfs etc., so I’m going to listen to them and see if I notice any differences and which I like better” is more advisable. 

[pkane2001]

18 minutes ago, JoshM said:

I followed the ASR attempt to establish agreed upon thresholds for a while, and it seems like it petered out. My sense of why that happened is that we can mean “audibility threshold” in two different senses: what humans physiologically can detect and what research has established listeners hear in audio reproduction. For example, the audibility threshold of distortion was stricter in the initial ASR threshold attempt than in Archimago’s recent experiment. So, over at ASR there seemed to be a mix of physiology, AES-type research, and random assertions used to support the “lenient” and “strict” thresholds. 
 

Some of the results, such as a “strict” linearity of 120 seemed far beyond realistic audibility IMHO. At the same time, I kind of think IMD is more audible than that ASR post suggested. However, it definitely seems like there’s a lack of good research on some topics.
 

Then, of course, we’d get into issues of age and hearing and trainer versus untrained listeners.

 

Ultimately, this uncertainty and complexity is why I think a lot of people default to the “moar is better” approach. But I think a “these DACs all clear the bar of flat FR, linearity to 90 dBfs etc., so I’m going to listen to them and see if I notice any differences and which I like better” is more advisable. 

 

The threshold thread was a bit convoluted and used data from mixed sources, including, as an example, the opinions posted by NwAvGuy. It was a good start, but it certainly did not define hard limits that would be acceptable to everyone, more like guidelines to get you started.

 

IMD/HD limits were recently tested by @Archimago in a blind test. A small percentage scored perfectly (not to brag, but I was one of them). 

 

Age-related hearing deterioration is absolutely something that should be tested if one is interested in audibility thresholds. A few years ago I wrote a quick-and-dirty program to let me measure my own hearing across the 20-20k frequency spectrum. Mine turned out to be worse than the ISO average by a small margin, but better than average for my age group.

 

One of the reasons I've been building software to simulate various levels of different types of distortions is to let me, and those who are interested, do the testing themselves. It doesn't help me if I know that someone out there can hear out past 20kHz if my hearing range is limited to 16k. I'd rather know my own thresholds than some published average or a statistical distribution from testing of some group.

 

 

[John Dyson]

2 hours ago, pkane2001 said:

 

This may be a good start, as a summary:

 

http://www.aes-media.org/sections/pnw/ppt/other/limitsofhearing.ppt

 

 

 

 

 

 

 

 

Wow -- there is something very interesting in the document that I found in the anti-distortion processing in the DHNRDS (REALLY!!!).  This correlation really validates (for me) a lot in the document, even though intellectually I agree with many other facts that I had previously known.

 

There is some anti-distortion code that has a cutoff that I found empirically to 'act differently' in the LF to 500-1kHz  range.  So, the processing is actually different, where the anti-md code SPECIFICALLY does not operate at all below apprx 750Hz, and there is an option to enable it between 650Hz to the 3000Hz range.  (it isn't really fully effective until about 1.5kHz though -- there is a slope in the filter curve starting at 650Hz.)   The anti-md (when enabled) is always effective between about 2.5kHz to 21kHz.

 

Point being -- I found that too much manipulation below about 1kHz causes a cringing kind of distortion -- and the gain control must match perfectly to the true HW.   Otherwise, above about 1kHz, the modulation effects become progressively more important- so the decoder eliminates most modulation of the signal.  Depending on the kind of distortion -- if it isn't of certain raw harmonic type, creates an imprecision in the sound more than a gritty/grainy sound.  Perfectly matching the HW above 1kHz isn't as important as just leaving the envelope alone when gain doesn't need to change, including minimizing the modulation effects.  (Also does a small amount of undoing previous modulation effects.)

 

Anyway, *independently*, I detected a difference below 1kHz and above 1kHz -- the AES document has a more scientific mention of that phenomenon.   Imagine how much frustrated whack-a-mole testing I did, and how much time I wasted using at-hoc techniques finding that cutoff?  (Answer:  A LOT.)

 

Loosing 'professional discipline' wasn't a good thing -- and one reason why I sometimes sound a little pedantic -- it is from experience.

 

Suggestion:  measure twice, cut once, not tweak tweak tweak.

 

John

 

[Speedskater]

5 hours ago, pkane2001 said:

This may be a good start, as a summary:

http://www.aes-media.org/sections/pnw/ppt/other/limitsofhearing.ppt

 

the Pacific Northwest Section of the Audio Engineering Society has a ton of great papers and Power Points:

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

 

by j.j. (James D. Johnston)

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

by other experts:

http://www.aes-media.org/sections/pnw/ppt/

[sandyk]

5 hours ago, The Computer Audiophile said:

If objectivists can agree that human adults can hear up to 20 kHz, they can surely agree on other limits when presented real data. 

 

 FWIW, Barry Diament's wife was able to hear 23kHz which is above CD's limitations .

[fas42]

What human hearing is very good at, is being able to focus on fine detail, amidst high levels of background or attendant sound - the cocktail party effect. Live music always has this quality, of there being "tiny things happening", while there is a very intense overall sound level - which we humans have little difficulty picking up. Unfortunately, generally audio rigs are poor at doing this, for a variety of reasons; which is why there is a common belief that playback can't mimic live music making - but, this failing is merely another form of distortion, which can be eliminated.

 

It would not be easy to simulate this, say in Paul's software, because it is a highly non-linear behaviour - being able to measure the levels of this type of distortion would be good, but I can't recall coming across any research looking at this - a common phrase of what people hear when a system does this is that the sound starts to compress; a capable replay doesn't do the latter, a common phrase for that is that the reproduction is effortless.

 

Currently, tweaking is the normal way to deal with this fault in typical audio replay; because the industry as a whole doesn't recognise this misbehaviour.

[The Computer Audiophile]

18 minutes ago, sandyk said:

 

 FWIW, Barry Diament's wife was able to hear 23kHz which is above CD's limitations .

FWIW, let’s not let the tail wag the dog. 

[Audiophile Neuroscience]

I could weigh in on audibility thresholds, differences for absolute vs differential and interesting results for  non periodic steady-state noise but as I see it, what the thresholds actually are is not the point.

 

The point is *if* someone declares a threshold for inaudibility (whether they be right or wrong), then uses it to further their opinion on another topic, but then elsewhere appears to use the very same data in a contradictory manner. A double standard is said to exist or at least is implied.

[yamamoto2002]

8 hours ago, pkane2001 said:

It doesn't help me if I know that someone out there can hear out past 20kHz if my hearing range is limited to 16k. I'd rather know my own thresholds than some published average or a statistical distribution from testing of some group.

 

You may be relieved to hear there is only Do, Do♯, Re and Re♯ between 16kHz to 20kHz

 

14,080Hz A    La
14,917Hz B(A♯)  La♯
15,804Hz H    Si
16,744Hz C    Do
17,739Hz C♯  Do♯
18,794Hz D    Re
19,912Hz D♯  Re♯
21,096Hz E    Mi
22,350Hz F    Fa
23,679Hz F♯  Fa♯
25,087Hz G    Sol
26,579Hz G♯   Sol♯
28,160Hz A     La

[pkane2001]

1 hour ago, yamamoto2002 said:

 

You may be relieved to hear there is only Do, Do♯, Re and Re♯ between 16kHz to 20kHz

 

14,080Hz A    La
14,917Hz B(A♯)  La♯
15,804Hz H    Si
16,744Hz C    Do
17,739Hz C♯  Do♯
18,794Hz D    Re
19,912Hz D♯  Re♯
21,096Hz E    Mi
22,350Hz F    Fa
23,679Hz F♯  Fa♯
25,087Hz G    Sol
26,579Hz G♯   Sol♯
28,160Hz A     La

 

That's good news! 

[John Dyson]

6 hours ago, The Computer Audiophile said:

FWIW, let’s not let the tail wag the dog. 

No matter what, some people hearing at 23kHz does kind-of argue for 48kHz sample rate.   Most 'music' doesn't have fundamentals above a few kHz, and by the time the harmonics on normal material have trailed off -- the acutal 'audio' above 12kHz is usually much attenuated by the time it gets into the 'brain'.

 

Detectability is very different than it being useful.  Audio based security systems and traffic light controllers in the olden days used to drive me totally nuts -- now I hear the buzzing all of the time :-).  On the other hand, I never heard music in the frequency range of the irritating super-high frequency audio sensors.

 

If someone can actually hear the 23kHz harmonics of real music audio, then I wouldn't want to be anywhere near the source.  I treasure my bones and don't want them to be pulverized.  You know, they become more and more brittle as we get older :-).

 

John

 

[John Dyson]

4 hours ago, Audiophile Neuroscience said:

I could weigh in on audibility thresholds, differences for absolute vs differential and interesting results for  non periodic steady-state noise but as I see it, what the thresholds actually are is not the point.

 

The point is *if* someone declares a threshold for inaudibility (whether they be right or wrong), then uses it to further their opinion on another topic, but then elsewhere appears to use the very same data in a contradictory manner. A double standard is said to exist or at least is implied.

I am all for a frequency response of a system that is as wide as reasonably possible.  44.1k is indeed cutting it close, but is okay for REAL music and not special effects.  I'd prefer the wiggle room of 48khz, and that is what I use for down-sampled material, even if starting with 88.2kHz.   I watch my spectograms very carefully -- MOST of the time, older pop music appears to contain modulated noise above 20kHz not so much actual information.  On the other hand, there ARE good recordings out there with short-term transients above 20kHz.  If there was significant provable information above 20khz on the material that I played with -- I'd use higher sample rates.   I vote for 66.15 and 72 as a compromise (that is what the DHNRDS uses for 44.1k and 48k material, BTW -- then does a further upconversion to 88.2k/96k.)    Nowadays, space is cheap, so it is kind of silly to consider 66.15k/72k except for very special purposes.

 

John

 

[John Dyson]

10 minutes ago, John Dyson said:

I am all for a frequency response of a system that is as wide as reasonably possible.  44.1k is indeed cutting it close, but is okay for REAL music and not special effects.  I'd prefer the wiggle room of 48khz, and that is what I use for down-sampled material, even if starting with 88.2kHz.   I watch my spectograms very carefully -- MOST of the time, older pop music appears to contain modulated noise above 20kHz not so much actual information.  On the other hand, there ARE good recordings out there with short-term transients above 20kHz.  If there was significant provable information above 20khz on the material that I played with -- I'd use higher sample rates.   I vote for 66.15 and 72 as a compromise (that is what the DHNRDS uses for 44.1k and 48k material, BTW -- then does a further upconversion to 88.2k/96k.)    Nowadays, space is cheap, so it is kind of silly to consider 66.15k/72k except for very special purposes.

 

John

 

I want to explain the modulated noise comment.  That come from the NR systems that they used to use.  It creates an envelope around hiss left over from the tape recorder when the HF bands change gain.  It is as simple as that.  When there are transients, they are such low level that they cannot be heard -- except by VERY SPECIAL PEOPLE.  This is especially true when decoding with DolbyA units -- long propagation in the decoding process -- misses transients.   The mics that used to be used (esp like U47) don't give you much above 20kHz, but they do peak between 6k-9k a little.  Ribbons were worse.   I had some ancient Altec condensers that did pretty well to 20kHz,  but they were omnis and very simple designs.   Earthworks mics didn't exist a long time ago, and DPA mics weren't all that common.   Normal condensers didn't appear to be designed for the max response.

 

There are now Earthworks mics, wider band RF condensers from Sennheiser, and always the DPA specialty mics -- but they are not likely used on common recordings.  The Sennhiser RF condensers are interesting though -- wideband AND low noise.  Earthworks, not so much -- just wideband.

 

John

 

[John Dyson]

6 hours ago, sandyk said:

 

 FWIW, Barry Diament's wife was able to hear 23kHz which is above CD's limitations .

But, when the levels from actual music material are below the hearing threshold anyway, who cares?  Super intense 23kHz isn't coming from any normal music, unless skin is being heated, nonlinear effects in the hearing system, or bones pulverized.   I used to hear security systems all of the time, probably in the low 20kHz range -- but music didn't happen up there (I could tell the difference, and always used high quality (but beamy) super-tweeters.)

 

What you usually see on older recordings done on nice, wideband tape, is noise modulation from the NR system...  That is it -- and, frankly, I don't care about noise modulation.  I worry more about the weakened/missed AUDIBLE & INADUBLE  transients from DolbyA/SR/DBX material.   Yes -- there is a VERY LONG delay through the decoding loop on all of the systems, (not so much C4) where transients are eweakened/missed, where most of the 20+kHz energy resides.   This is one of the major reasons why the sound of cymbals is so attenuated on even properly mastered material.

 

John

 

[John Dyson]

Deleted -- about filter choices messing up the comparisons about sample rate conversions -- but kind of off topic.

 

[pkane2001]

49 minutes ago, John Dyson said:

I have been thinking (honestly) about the matter of comparing sample rates, and how people DO sometimes hear differences.  Was thinking about errors in the experiments, and one error that seems to be very easy to make -- using not-linear phase filters in the conversions.  Let me explain:

 

The difference between say, minimum phase and linear phase isn't the wiggles that move around on square waves, but more the delay at low frequencies is very different than high frequencies on a minimum (or intermediate) phase filter.  This means that significant timing differences encroach into the audible spectrum.   The ONLY kind of filter where the delays are constant and the signal will remain maximally unmodified is a constant delay filter.  That means, 'linear phase'.

 

I haven't looked at any experiment in detail, because there are usually far more details missing than let any real evaluation be done.   As soon as 'software in a box' is used, I am not 100% sure that I trust its technical accuracy.   Anyway -- that is off topic,  it seems like there is a common 'tone control' (:-)) used by some audio people, choosing between linear phase, minimum phase and intermediate phase filter regimes.

 

An experiment about sample rates cannot be very consistent without using constant delay filters.   That doesn't mean, when all is said and done, the 'filter of choice' wont be used after the experiment.  But as soon as the delays are variable, then the results are in question when comparing the audibility of sample rates.   Again, after the experiment, use the filter choices that 'sound good' -- but the only way to do the experiments with not-linear-phase filters is to make sure that the delays are consistent between filter choices.

 

John

 

I didn't score well on the HD-Audio Challenge by Mark Waldrep, as an example. But I know a few others who scored perfectly, so there's something related to hi-res encoding or its playback that can possibly make these audible. I doubt that it has anything to do with the frequency response caused by the increased sampling rate between 44.1kHz and 96kHz. But then, the question is what is it? Is it the filter? The resampling in the DAC or the reconstruction filter? IMD with higher-frequency signals? Or are some ears just more sensitive to it than mine (I've no doubt that's true)? The reports of someone having useful hearing at 23kHz as an adult are possible, but very unlikely.

 

Here is a study result based on 384 test subjects of various ages. You'll note that the level of sound at 20kHz needs to be over 90dB to be detectable for age group 22-35 (and no, I'm not even remotely close to that age group!) The error bands go down to about 85dB level, so not much variation. I don't think I'd ever want to listen to a recording that had 90dB content above 20kHz:

 

[John Dyson]

29 minutes ago, pkane2001 said:

 

I didn't score well on the HD-Audio Challenge by Mark Waldrep, as an example. But I know a few others who scored perfectly, so there's something related to hi-res encoding or its playback that can possibly make these audible. I doubt that it has anything to do with the frequency response caused by the increased sampling rate between 44.1kHz and 96kHz. But then, the question is what is it? Is it the filter? The resampling in the DAC or the reconstruction filter? IMD with higher-frequency signals? Or are some ears just more sensitive to it than mine (I've no doubt that's true)? The reports of someone having useful hearing at 23kHz as an adult are possible, but very unlikely.

 

Here is a study result based on 384 test subjects of various ages. You'll note that the level of sound at 20kHz needs to be over 90dB to be detectable for age group 22-35 (and no, I'm not even remotely close to that age group!) The error bands go down to about 85dB level, so not much variation. I don't think I'd ever want to listen to a recording that had 90dB content above 20kHz:

 

One thing that I'll do when I get a chance, actually create a table of delay vs frequency for various minimum phase filters -- I am truly not sure the effect, but the difference between DIFFERENT minimum phase filters can create audible differences (theoretically.)

 

My guess that the differences that people hear, and whether or not they hear differences is based upon time resolution of the hearing at audible frequencies, not so much the higher sample rate/Nyquist frequency per se.  Also, don't mistake my comment about sample rate -- it has little to do with the time resolution, as long as there is enough bit resolution.  (It is all about maximum information content.)

 

There is NOTHING wrong with using minimum phase filters if that is what 'sounds good', but for an experiment -- but I believe that filter delay vs. frequency will easily bias the results.   I would ONLY use carefully crafted software for testing purposes also -- not something 'off the shelf.'.  The reason isn't that other DSP software developers aren't competent, but on the other hand, the design might not have considered things that are important in the experiment.

 

I remember a kid when I was in high school was passing hearing tests at impossibly low levels because he was hearing the hiss in the electronics as it was gating on and off.  These measurements must be done with a scientific mentality, not just scientific/good experimental discipline.   As I tell everyone on some of these things 'There be dragons'.

 

John