Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Next page in the hi-rez media scam: A Meta-Analysis of High Resolution Audio Perceptual Evaluation (Read 96213 times) previous topic - next topic
0 Members and 2 Guests are viewing this topic.

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #50
those arguments are weak. Only last cca 20 years  we have real technology that can do 24 bit and/or 48/96 kHz sample rates reasonably well. So no century wide research ....
Believers believe we've only been able to generate >20khz signals for 20yrs??

If you set the bar high enough, you can push forward the point in time when a certain situation was first true.

For example, the AES Audio Timeline

http://www.aes.org/aeshc/docs/audio.history.timeline.html

says "1996 - Experimental digital recordings are made at 24 bits and 96 kHz."

I know that I was able to make high quality 24/96 recordings using B&K omni mics with flatish response up to 40 KHz and a reasonably priced pro audio recording interface card  (Card Deluxe)  ca. Y2k.

I doubt that there are many who are aware of the various work-arounds that were used to gather experimental data > 20 KHz before then.

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #51

I do not want to strive for Hi-Res at all costs, but those arguments are weak.

I don't expect to convert any true believers in high resolution audio with logical arguments.

Quote
Only last ca 20 years  we have real technology that can do 24 bit and/or 48/96 kHz sample rates reasonably well. So no century wide research ....

I guess you haven't noticed that we still can't do 24 bits.

More to the point, there is no practical purpose that would be satisfied by being able to do so.  In fact 16/44 is an overkill format as compared to the limitations of human hearing and musical events.

If you want to see an example of a weak argument, consider the argument that after having 24/96 gear with high performance at our disposal for 20-ish years, the best that high resolution advocates can come up with seems to be results that are less than 4% better than Placebo.


Interesting. Just like to add that I know that SNR of most equipment is between 100-120 dB, e.g. not fully utilizing 24 bit 144 dB range. But there is also quantization error. Considering sample rates isnt 48 kHz enough for very good filtering? And it is not about generating >20kHz signal but widespread availability of 24 bit and/or 48/96 Khz DACs.

Do not want to flame about Hi-Res, BTW. That discussion is endless, but I still feel that here too restrictive approach prevails, although i fully respect that up to now we do not have any prove that it is neccessary for complete listening experience.

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #52
Only last ca 20 years  we have real technology that can do 24 bit and/or 48/96 kHz sample rates reasonably well. So no century wide research

And it is not about generating >20kHz signal but widespread availability of 24 bit and/or 48/96 Khz DACs.
::)
Loudspeaker manufacturer


Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #54
...although i fully respect that up to now we do not have any prove that it is neccessary for complete listening experience.
This almost made me giggle.

Sane people would probably say that we have proof enough that HRA isn't necessary for a complete listening experience. But that would require a reasonable notion of what constitutes a good enough proof, and what constitutes a good enough approximation of a complete listening experience.

But that's unlikely to get accepted by audiophiles. For them, there can't ever be a proof convincing enough, and an experience complete enough. That's a matter of principle.

Heck, I'm trying to work out in my head what a 'complete listening experience' would actually be. It certainly wouldn't circle around getting presented with frequencies my ears can't hear, in the hope of me being able to perceive them in some yet unexplained way. My associations are much more in the direction of getting presented the complete soundfield rather than only a stereo approximation, and with getting all the non-acoustic stuff that contributes to an experience. I'm human after all, and I perceive through all my senses, and all this contributes to an experience.

So my answer to those wanting a complete listening experience would be: Forget it. HRA won't get you a micron closer, since the problem is somewhere entirely different. This entire HRA bandwagon is barking up the wrong tree. And for a long time already.

 

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #55

Interesting. Just like to add that I know that SNR of most equipment is between 100-120 dB, e.g. not fully utilizing 24 bit 144 dB range.

If you include mics and mic preamps as "equipment", then 100 dB is probably the more realistic number.

Quote
But there is also quantization error

Just another influence that is about 100 dB down

Furthermore, with perceptual noise shaping, it can be the perceptual equivalent of flat noise shaped  quantization noise 120 dB down

Quote
Considering sample rates isnt 48 kHz enough for very good filtering?

Filtering only needs to be good enough to avoid audible artifacts.  Done deal @ 44 KHz.  You do have to pay a little attention to what you are doing and avoid doing things the worst possible way.

Quote
And it is not about generating >20kHz signal but widespread availability of 24 bit and/or 48/96 Khz DACs.


That's off topic. The title of the paper is: A Meta-Analysis of High Resolution Audio Perceptual Evaluation. Notice that ADCs and DACs are not the issue, but what people can hear is.

Now that we've cleared up that little problem, we see that the question is all about generating > 20 KHz signals by any of a number of reasonable means.

This is one of those places where we separate the scientists from the audiophiles. The audiophiles think of problems in terms of specific pieces of audio gear. Scientists think about things like first principles and basic physical processes.



Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #56
I fully agree with the point of 24 vs 16 bit, but I think 44KHz vs 192KHz is less clear cut, and is highlighted perfectly by the pono player (though its digital filters). Playing back 192KHz the filters are less-likely to screw up the actual stuff you can hear, so in many respects 192KHz has the advantage.

So that means that if a create a device with good filters at 44.1 kHz and horrible filters at 192 kHz, it's a reasonable argument that 192 kHz is inferior, right?

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #57
The misleading point of this last AES paper is that it summarizes a trained listener can distinguish High Samplerates even when letting out what this means. It repeats what people like BS wanted to spread with own papers.
Every audiophile absolutely feels himself as well trained now and all that can't hear it are not deaf but ignorant.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #58
The misleading point of this last AES paper is that it summarizes a trained listener can distinguish High Samplerates even when letting out what this means. It repeats what people like BS wanted to spread with own papers.
Every audiophile absolutely feels himself as well trained now and all that can't hear it are not deaf but ignorant.
Yes, I posited this dilemma in the AES comments section. Training for hearing what?
Loudspeaker manufacturer

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #59
Also many of these former papers had flaws that are completely left out. Alone the BS paper even when positive to blame filter ringing to be audible needs very strong additional ringing in a steep 192->44.1->192 resampling not present on normal distribution.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #60
I fully agree with the point of 24 vs 16 bit, but I think 44KHz vs 192KHz is less clear cut, and is highlighted perfectly by the pono player (though its digital filters). Playing back 192KHz the filters are less-likely to screw up the actual stuff you can hear, so in many respects 192KHz has the advantage.

So that means that if a create a device with good filters at 44.1 kHz and horrible filters at 192 kHz, it's a reasonable argument that 192 kHz is inferior, right?

I suppose it is possible but less likely, you see a device filtering around the 22KHz range is very close to what you can hear, a 192KHz playback device which filters above 60KHz for example is far above the audible range.

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #61
The point I was (indirectly) making is we need to keep clear whether we are talking about problems with 44.1 kHz in general or with a specific device/DAC/filter's implementation at 44.1 kHz.

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #62
Playing back 192KHz the filters are less-likely to screw up the actual stuff you can hear, so in many respects 192KHz has the advantage.
I guess you haven't seen this, which essentially amounts to a prerequisite for this community:
http://xiphmont.livejournal.com/58294.html

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #63
I have seen that many years ago, and it is well know that almost all audio DACS over sample to precisely create some headroom for the filter to work. Which is why the pono is so puzzling.

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #64
I suppose it is possible but less likely, you see a device filtering around the 22KHz range is very close to what you can hear
Well, if one is 12yrs old. I'm 50 and struggling by 15KHz.
I'd wager old deaf audiophiles like Neil Young driving this craze are done by 11KHz. Not quite very close IMO.
But I believe the phantom menace here is that dreaded "time smear"....
Loudspeaker manufacturer

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #65
No one is questioning the validity of oversampling as a means to help eliminate imaging; rather, this is about whether 44.1k as a delivery format is adequate for reproduction of recorded material.

There are plenty of hardware devices that get it wrong.

There have been countless discussions on these things going all the way back to this forum's infancy.

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #66
The misleading point of this last AES paper is that it summarizes a trained listener can distinguish High Samplerates even when letting out what this means. It repeats what people like BS wanted to spread with own papers.
Every audiophile absolutely feels himself as well trained now and all that can't hear it are not deaf but ignorant.
Yes, I posited this dilemma in the AES comments section. Training for hearing what?

Looking at the results in the paper that made it though the author's cherry-picking. It appears that there were 12,645 trials half A and half B. If the listeners were guessing purely randomly, they would have obtained correct identification 6, 3,22 times.  They did a trifle better than that and obtained correct identification 6, 736 times.  IOW 414 times out of 12,545 trials or about once each 30 trials the listeners  actually heard a difference.  The author seems to have provided no information about the training nor any kind of indication what the training actually accomplished.

I'll give the author the benefit of the doubt, and estimate that the listeners  accuracy was doubled. IOW they now provided a correct response for every 15 trials. Trouble is, training people to do something that they fail so often is very difficult and frustrating. 

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #67
The author didn't perform the test, but is rather sifting through the data of
http://www.aes.org/e-lib/browse.cfm?elib=17497
among other tests.

I am certain he has no answer, besides the point that AJ's question was rhetorical.  I'm afraid that only our friends at Meridian and the people who were involved in the process know the answer.  I wouldn't hold my breath waiting for any further information.

Like I previously intimated, they have their slam dunk and are now busy selling licenses for MQA.

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #68
have seen that many years ago, and it is well know that almost all audio DACS over sample to precisely create some headroom for the filter to work.

The primary  purpose of oversampling relates to improving dynamic range. 

It is very feasible to have a digital filter that is very effective and also does not use oversampling.  IOW it operates at the identical same clock frequency as the data it processes.

It is also possible to obtain good measured and audible performance out of brick wall filters that are entirely implemented in the analog domain, it is just that they are very expensive to make properly.  There were used in the first several generations of digital audio recorders such as the Sony PCM 1610[

http://www.realhomerecording.com/docs/Sony_PCM-1610_brochure.pdf

There is about 6 KHz between the point where a sharp low pass filter "brickwall" ceases to have audible effects and the Nyquist frequency associated with 16/44 digital.  Its not all that tight.

Quote
Which is why the pono is so puzzling.

They are just taking advantage of the fact that digital filters can be designed to be sonically transparent or not, depending on the will of the designer. The Pono is not unique in possessing digital filters with audible effects.  Its the same basic philosophy as SET power amps.

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #69
Furthermore, with perceptual noise shaping, it can be the perceptual equivalent of flat noise shaped  quantization noise 120 dB down.
Even without taking perceptual noise shaping into account, the oft-repeated argument that human hearing can span a dynamic range of up to 120 dB, hence we need 20 bit, is misguided. Reiss also doesn't fail to mention this flawed argument:
Quote
It is well-known that the dynamic range of human hearing (when measured over a wide range of frequencies and considering deviations among subjects) may exceed 100 dB. Therefore, it is reasonable to speculate that bit depth beyond 16 bits may be perceived.
The flaw with this argument is in two different and incompatible notions of dynamic range, i.e. it is a case of comparing apple with orange:

In digital audio, the dynamic range is defined as the difference between a full scale sine wave and the noise floor. Note that a sine wave is being compared with noise, which makes it dubious already.

In human hearing, dynamic range is defined as the difference between the loudest and the softest sine wave that can be heard. Hence the frequency dependency, i.e. Fletcher-Munson. Note that there's no mentioning of noise here. Of course the loudest sine wave is a weakly defined term, since there is no hard limit. You have to put the limit somewhat arbitrarily, depending on the damage level and distortion you are willing to accept.

It should be well known that human hearing can hear sine waves buried in the noise floor. That's not as astonishing as some would think, because any frequency-selective measurement can do the same. It is all about measurement bandwidth. Where the ear has its bark bands, a measurement instrument could use a filter bank, or use FFT. The result is that the noise floor as calculated from the number of bits is not the limit of audibility for a tone. Hence you can't apply digital audio wordlengths directly to Fletcher-Munson curves.

If one wanted to do a fair comparison, one would have to relate the noise floor of digital audio with the noise floor of the human ear. You can't glean that from the Fletcher-Munson curves. The picture would be rather different. You'd suddenly find that (surprise!) 16 bits are sufficient.

I have seen that many years ago, and it is well know that almost all audio DACS over sample to precisely create some headroom for the filter to work.
It would be better to say: ...to create some headroom to make the analog filter simpler.

It's not that without oversampling the filter wouldn't work. Numerous implementations show that it would, given enough care in designing it. Oversampling is a way of putting some of the reconstruction work into the digital domain, so that less of it needs to be done in the analog domain. The cheaper digital circuitry becomes, compared to analog circuitry that does the same job, the more economic sense oversampling makes. Given the fast advance in digital technology, you shouldn't be surprised that over time, the economic advantage of digital technology became greater and greater, so the push to do more and more of the reconstruction on the digital side got bigger and bigger. The sigma-delta technology is the pinnacle of this. You can today get converter chips for a dollar which contain everything except a passive RC-filter needed on the output, which produce a fidelity that you would never be able to match with a filter that's 100% in the analog domain.

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #70
Furthermore, with perceptual noise shaping, it can be the perceptual equivalent of flat noise shaped  quantization noise 120 dB down.
Even without taking perceptual noise shaping into account, the oft-repeated argument that human hearing can span a dynamic range of up to 120 dB, hence we need 20 bit, is misguided.

Agreed.  The oft-quoted source for  measurements showing that the ear has > 120 dB dynamic range is often " Louis D. Fielder. May 1981. Dynamic Range Requirement for Subjective Noise Free Reproduction of Music"  That argument and others are based on the idea that the human ear can hear a pure tone at a frequency where it is most sensitive with an amplitude of about - 6 dB. The threshold of pain is usually given as 120 dB SPL or so, so the ear's dynamic range must be ca.  126 dB and we need reproduction systems capable of doing this.

Engaging the real world points out a number of flaws when it is applied to recordings of real world that are listened to for pleasure.

The first problem I've noticed is that listening to very loud music causes the ear to experience short and intermediate term insensitivity, and listening to music that peaks in the 110 dB and greater for any but the briefest periods of time range desensitizes the ear to the point where hearing normal speech (ca. 60 dB SPL) , let alone sounds near the threshold of hearing, can be difficult or impossible.  If you use these experiences to estimate the dynamic range of the human ear, it is about 60 dB.

Careful reading of Fielder's paper shows that his reference sounds for judging  loud music was frequently based on electronically amplified drums.  Years of experience doing technical setup and mixing for live concerts with typical electronically-amplified instruments revealed to me  the fact that the noise floor of the electronic equipment sound reinforcement equipment being used is far from being SOTA, and sometimes clearly audible in the paying seats during the concert.   The peak levels may be > 120 dB SPL, but the noise floor may again  be > 60 dB.

Measuring the loudest sound that can be heard one day to the softest sound that may be heard on some other day in some other place presumes that the device being measured is free of nonlinear effects such as intermodulation distortion, dynamics compression and unaffected by the sounds being listened to no matter how loud. Those must be some golden ears because I have never encountered them in real life.

Quote
Reiss also doesn't fail to mention this flawed argument:
Quote
It is well-known that the dynamic range of human hearing (when measured over a wide range of frequencies and considering deviations among subjects) may exceed 100 dB. Therefore, it is reasonable to speculate that bit depth beyond 16 bits may be perceived.
The flaw with this argument is in two different and incompatible notions of dynamic range, i.e. it is a case of comparing apple with orange:

In digital audio, the dynamic range is defined as the difference between a full scale sine wave and the noise floor. Note that a sine wave is being compared with noise, which makes it dubious already.

In human hearing, dynamic range is defined as the difference between the loudest and the softest sine wave that can be heard. Hence the frequency dependency, i.e. Fletcher-Munson. Note that there's no mentioning of noise here. Of course the loudest sine wave is a weakly defined term, since there is no hard limit. You have to put the limit somewhat arbitrarily, depending on the damage level and distortion you are willing to accept.

It should be well known that human hearing can hear sine waves buried in the noise floor. That's not as astonishing as some would think, because any frequency-selective measurement can do the same. It is all about measurement bandwidth. Where the ear has its bark bands, a measurement instrument could use a filter bank, or use FFT. The result is that the noise floor as calculated from the number of bits is not the limit of audibility for a tone. Hence you can't apply digital audio wordlengths directly to Fletcher-Munson curves.

If one wanted to do a fair comparison, one would have to relate the noise floor of digital audio with the noise floor of the human ear. You can't glean that from the Fletcher-Munson curves. The picture would be rather different. You'd suddenly find that (surprise!) 16 bits are sufficient.

Agreed.

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #71
The first problem I've noticed is that listening to very loud music causes the ear to experience short and intermediate term insensitivity, and listening to music that peaks in the 110 dB and greater for any but the briefest periods of time range desensitizes the ear to the point where hearing normal speech (ca. 60 dB SPL) , let alone sounds near the threshold of hearing, can be difficult or impossible.  If you use these experiences to estimate the dynamic range of the human ear, it is about 60 dB.

Not to mention the permanent hearing damage associated with listening to music at these sound pressure levels. Unless that's what you meant by desensitizing?

I've certainly been a lot better at remembering my earplugs since:

1) I developed tinnitus, which sucks. Young people, wear your goddamn earplugs!
2) I had my hearing tested and discovered hearing loss around 3-4kHz and above ~14kHz on my left ear and above ~16kHz on my right ear (and I'm only 30yo)
3) I happened to stand beside the mixing desk at an indoor concert and noticed that the normal sound level was around 105dB with peaks all the way up to 120dB. I hadn't really thought about it before.

Most of my friends don't wear earplugs to concerts, and I expect them to be almost completely deaf around the age of 50 or so. A lot them have the attitude that "if it's too loud, you're too old", but I think it should be the other way around.

As an aside, most audiophiles who claim to be able to hear super-high frequencies tend to be >50yo, so it's just painfully obvious that they're just bullshitting.

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #72
The oft-quoted source for  measurements showing that the ear has > 120 dB dynamic range is often " Louis D. Fielder. May 1981. Dynamic Range Requirement for Subjective Noise Free Reproduction of Music"  That argument and others are based on the idea that the human ear can hear a pure tone at a frequency where it is most sensitive with an amplitude of about - 6 dB. The threshold of pain is usually given as 120 dB SPL or so, so the ear's dynamic range must be ca.  126 dB and we need reproduction systems capable of doing this.
I reviewed this article, and I think you remember it wrongly. Fielder actually refers to just noticeable noise levels, rather than just noticeable tone levels. So he actually does compare apples to apples.

In order to reach 118 dB of required dynamic range, he had to take the most extreme percussive classical music, close-mike it, and assume the most acute listener in detecting a noise floor increase. You'd have to have the very quietest part, where noise floor differences might be heard by a few people, before the loud part, because after the loud part nobody would detect such noise floor differences anymore.

But since he assumed dither with no noise shaping, and since the sensitivity to background noise is best between 3 and 7 kHz, you still have the possibility to use noise shaped dither to give you the desired dynamic range with 16 bit, even though the case is already unrealistic and extreme.

If Fielder's work is used to justify 20 bit or 24 bit for professional gear used in recording and production, I'm all for it. But that's what we have today, anyway. As an argument for HRA for distribution to consumers, it doesn't fit. He doesn't say so, either, so I don't blame him.

You can get the dynamic range required for all realistic cases from the CD if you want to, and if you know what you are doing. This would even apply when you really wanted to release a disk with no compression at all, which hardly anybody does, even for classical concerts.

In this sense, I agree with you.

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #73
"Results showed a small but statistically significant ability of test subjects to discriminate high resolution content,
and this effect increased dramatically when test subjects received extensive training."
That is the contention which originated from Meridian, perhaps even a verbatim talking point.  Don't be fooled into thinking otherwise.

Anyway, like the hypersonic mumbo jumbo, there has been zero 3rd party verification.

Seriously, the discussion could just end now, but I don't think that would make krab happy.

Who, me?

Meta analysis is of course extremely dependent on the corpus of work used in the analysis.  What is included, what is left out, what assumptions are made.

Dr. Reiss does seem to have the background to understand statistics (PhD physics, with a focus on chaotic events).  But he does seem to have missed some of the literature though, given that he thinks his discussion of Type II errors in audio testing is a new thing when it's not

His take on Meyer and Moran is irritating, though (for the most part*) hardly novel.  Basically 'not sciency enough', but then again, M&M were merely allowing golden ears to do what they always do when they claim to hear 'hi rez' magic, while adding a blinding step.  Using the same kinds of recordings already reported to be 'magic' by them.  So who cares if the recordings  didn't actually have hi rez content?

My response to the overall 'implication' of this MA -- that in rare cases, some small number of people with documented training appear to be  hearing 'something' -- remains what it always has been to such 'findings':  so f*cking what?  That finding is NOT what audiophile/high end mavens and cheerleaders claim.  They typically say the difference is 'obvious', 'my wife could hear it' etc.  They consider themselves to be 'self trained', and use patently useless methodology.    Dr. Reiss gives nary a  word to that, despite the near certainty that Stereophile. Bob Stuart,  and the rest of the cheering squad will tout these results without the qualifications required by science.

My other take on these results would be  along the lines of his discussion paragraph, where he basically says: more replication of 'interesting results' is needed.  Though he seems to assume that more rigorous work would merely strengthen the implication of his meta-analysis.  I'm not so sure.



*I was gobsmacked to see him recite this argument though:  " the encoding scheme on SACD obscures frequency components above 20 kHz and the SACD players typically filter above 30 or 50 kHz"
So THAT's a reason now  why M&M wasn't a good test of audiophile claims? Give me a f*cking break!

Re: A Meta-Analysis of High Resolution Audio Perceptual Evaluation

Reply #74
Furthermore, with perceptual noise shaping, it can be the perceptual equivalent of flat noise shaped  quantization noise 120 dB down.
Even without taking perceptual noise shaping into account, the oft-repeated argument that human hearing can span a dynamic range of up to 120 dB, hence we need 20 bit, is misguided.

Agreed.  The oft-quoted source for  measurements showing that the ear has > 120 dB dynamic range is often " Louis D. Fielder. May 1981. Dynamic Range Requirement for Subjective Noise Free Reproduction of Music"  That argument and others are based on the idea that the human ear can hear a pure tone at a frequency where it is most sensitive with an amplitude of about - 6 dB. The threshold of pain is usually given as 120 dB SPL or so, so the ear's dynamic range must be ca.  126 dB and we need reproduction systems capable of doing this.

It may be oft-quoted  but it was superseded by Fielder 1994

http://www.aes.org/e-lib/browse.cfm?elib=10206

free download: 
http://www.aes.org/e-lib/inst/download.cfm/10206.pdf?ID=10206