Audio Science: Does it explain everything about how something sounds?

[amirm]

Here is some food for thought...

Let's talk lossy audio compression like MP3/AAC. No one uses any of traditional audio measurements to assess their performance. Everything is done with listening tests.

But, that is not the full picture. A 128 kbps file has 94% of the original data thrown away from the CD rip and you are listening to just 6% of it! Let's agree that to vast majority of people, that 6% essentially equals the original. No intuitive logic would lead you to such an outcome. Can you take out 94% of what I am typing and if it essentially be indistinguishable to what you started with? Clearly not.

Inside the lossy encoder there is a perceptual model of our hearing. Audio is divided into frames of X milliseconds and analyzed and different decisions are made in how the resolution of each frequency band is severely reduced while making sure the resulting distortions are below audible thresholds. It is a testament to audio research and how much we know about our hearing system that allows these systems to work so well.

Even at higher bit rates of 320 kbps, we are throwing out 75% of the audio. Despite that, most audiophiles will fail to detect a difference between those files and originals. The system is that good, and the knowledge of our hearing that capable.

But again, we perform listening tests because the model is not perfect. And ultimately we want the validation of a human saying something sounds good, or sounds better than an alternative.

We have an entire field called psychoacoustics which quantifies what we hear. It is through that science that enables things like MP3 to work. We also have powerful signal processing knowledge such as dither that applies noise to a single, something a lay person would think is "wrong," yet can be shown mathematically to reduce distortion.

One can't measure how much the research and industry know, by utilizing knowledge gained as a hobbyist. At least I never got exposed to it that way. I had to manage the research and development teams that did the work.

 

[Atmasphere]

Here is some food for thought...

Let's talk lossy audio compression like MP3/AAC. No one uses any of traditional audio measurements to assess their performance. Everything is done with listening tests.

But, that is not the full picture. A 128 kbps file has 94% of the original data thrown away from the CD rip and you are listening to just 6% of it! Let's agree that to vast majority of people, that 6% essentially equals the original. No intuitive logic would lead you to such an outcome. Can you take out 94% of what I am typing and if it essentially be indistinguishable to what you started with? Clearly not.

Inside the lossy encoder there is a perceptual model of our hearing. Audio is divided into frames of X milliseconds and analyzed and different decisions are made in how the resolution of each frequency band is severely reduced while making sure the resulting distortions are below audible thresholds. It is a testament to audio research and how much we know about our hearing system that allows these systems to work so well.

Even at higher bit rates of 320 kbps, we are throwing out 75% of the audio. Despite that, most audiophiles will fail to detect a difference between those files and originals. The system is that good, and the knowledge of our hearing that capable.

But again, we perform listening tests because the model is not perfect. And ultimately we want the validation of a human saying something sounds good, or sounds better than an alternative.

We have an entire field called psychoacoustics which quantifies what we hear. It is through that science that enables things like MP3 to work. We also have powerful signal processing knowledge such as dither that applies noise to a single, something a lay person would think is "wrong," yet can be shown mathematically to reduce distortion.

One can't measure how much the research and industry know, by utilizing knowledge gained as a hobbyist. At least I never got exposed to it that way. I had to manage the research and development teams that did the work.

Seriously, let's not talk about lossy compression. Its not hifi to begin with. It is the mark of a good engineer to know what is 'negligible' and what is not. lssy cmprssn isn't negligible. Its a crime.
I find it irritating if I know the music as stuff that I expect to hear isn't there. It sucks, yet based on the idea of the masking principle, which we know to be true, but you can still hear that it doesn't work. I don't know where the disconnect is- appication of a faulty algorithm or what, but it really doesn't matter. It sucks anyway you look at it and IMO its why people say they don't really care enough about music to make an investment.

IOW you don't have to know anything about the technical side- just the economics of what is happening not just to high end but also the music industry.

This BTW is the same problem with the tubes/transistor debate. You don't have to know the technical (who is right or wrong) just look at the market. If tubes really were obsolete they would have been gone 60 years ago. That they aren't says that they are not obsolete. Usually obsolete technology (like side valves in an engine) simply disappear from the market...

 

[jkeny]

Fair point.

However, different test signals to what are currently used, which take into consideration psychoacoustic measurements, still don't mean that we cannot measure everything about electrical/sound waves -- which is what the 'measurement objectivists' are all about. It's just the manner we go about those measurements, and the question what they all mean for human perception, that are at stake here.

But, in a way, isn't that like saying that we had the technology decades ago to find the evidence for the existence of the Hadron composite elementary particle?
People may have said that in those decades but it isn't really very helpful, is it? It required a great deal of work to use a combination of existing technologies over the decades to develop the large Hadron collider & the sophisticated analysis necessary to sense the traces of the existence of this particle.

And remember this whole test system was developed specifically because there was a hypothesis that could be tested - in other words the hypothesis of the deep structure of matter gave rise to logical predictions which were then tested in the Hadron collider.

Where are the hypotheses for audio reproduction & psychoacoustics that allow us to develop experiments to falsify/confirm the hypothesis? Without that we can't say that we can measure anything - it's just a bland, meaningless statement. Fact of the matter is that 'measurement objectivists' tend to be engineering driven, not research driven - they are not accustomed to thinking in these terms - they tend to stick to existing measurements & existing limits - this is their self-contained worldview

 

[jkeny]

Here is some food for thought...

Let's talk lossy audio compression like MP3/AAC. No one uses any of traditional audio measurements to assess their performance. Everything is done with listening tests.

But, that is not the full picture. A 128 kbps file has 94% of the original data thrown away from the CD rip and you are listening to just 6% of it! Let's agree that to vast majority of people, that 6% essentially equals the original. No intuitive logic would lead you to such an outcome. Can you take out 94% of what I am typing and if it essentially be indistinguishable to what you started with? Clearly not.

Inside the lossy encoder there is a perceptual model of our hearing. Audio is divided into frames of X milliseconds and analyzed and different decisions are made in how the resolution of each frequency band is severely reduced while making sure the resulting distortions are below audible thresholds. It is a testament to audio research and how much we know about our hearing system that allows these systems to work so well.

Even at higher bit rates of 320 kbps, we are throwing out 75% of the audio. Despite that, most audiophiles will fail to detect a difference between those files and originals. The system is that good, and the knowledge of our hearing that capable.

But again, we perform listening tests because the model is not perfect. And ultimately we want the validation of a human saying something sounds good, or sounds better than an alternative.

We have an entire field called psychoacoustics which quantifies what we hear. It is through that science that enables things like MP3 to work. We also have powerful signal processing knowledge such as dither that applies noise to a single, something a lay person would think is "wrong," yet can be shown mathematically to reduce distortion.

One can't measure how much the research and industry know, by utilizing knowledge gained as a hobbyist. At least I never got exposed to it that way. I had to manage the research and development teams that did the work.

Yes, Amir, we have advanced quite a long way in audio masking with fairly simple psychoacoustic models of auditory perception. It's a pity that this knowledge was directed towards bandwidth reduction rather than other audio improvement - in other words the focus was on how much could be removed without being too noticeable rather than trying to use knowledge of psychoacoustics to analyse existing audio reproduction & analyse it's shortcomings from a psycho-acoustical perspective!

Guess that's not where the funding was directing research & that's the biggest problem to advancing audio reproduction - there's no commercial gain to be had by any of the big organisations that have the funds to do this sort of research.

 

[FrantzM]

Seriously, let's not talk about lossy compression. Its not hifi to begin with. It is the mark of a good engineer to know what is 'negligible' and what is not. lssy cmprssn isn't negligible. Its a crime.
I find it irritating if I know the music as stuff that I expect to hear isn't there. It sucks, yet based on the idea of the masking principle, which we know to be true, but you can still hear that it doesn't work. I don't know where the disconnect is- appication of a faulty algorithm or what, but it really doesn't matter. It sucks anyway you look at it and IMO its why people say they don't really care enough about music to make an investment.

IOW you don't have to know anything about the technical side- just the economics of what is happening not just to high end but also the music industry.

This BTW is the same problem with the tubes/transistor debate. You don't have to know the technical (who is right or wrong) just look at the market. If tubes really were obsolete they would have been gone 60 years ago. That they aren't says that they are not obsolete. Usually obsolete technology (like side valves in an engine) simply disappear from the market...

What about if you don't know? Would you reliably detect the absence? Further would most people? Even those who consider lossy compresion a crime... For the recordthe audiophile in me even when I can't fully recognize mp3 from lossless prefer lossless
Don't laugh people do yourselves a favor and compare with knowledge removed mp3 320 to full CD and see how good you are in determining whihc is which without prior training.

 

[DonH50]

A lot of the measurement problem is making the right test signals. I wonder what is done in the way of impulse, modulated tones, broadband noise tests (ACPR-type tests), square waves, modulated square waves, and the like during design phases. So much that we could measure (or are measuring) is not reported and correlated to listening. Frequency response, basic THD/SNR/SINAD, and power output is a pitiful small sampling of the measurement suite we can and should be doing. Some of that may be because there is no real incentive to do so, and some of it may be because they are not standardized. And all the usual reasons (e.g. nobody wants to be first to report a new spec; the competitors will then provide a slightly better number, natch).

 

[Al M.]

Fact of the matter is that 'measurement objectivists' tend to be engineering driven, not research driven - they are not accustomed to thinking in these terms - they tend to stick to existing measurements & existing limits - this is their self-contained worldview

Here you bring up a sore point. Engineers are so cock-sure about their methods, like scientists would never be. Scientists know that, the more we know, the more we know what we don't know.

 

[Tony Lauck]

Seriously, let's not talk about lossy compression. Its not hifi to begin with. It is the mark of a good engineer to know what is 'negligible' and what is not. lssy cmprssn isn't negligible. Its a crime.

CD itself is a lossy encoding method. It throws away information that is in the original sound (if analog) or high resolution master. If the master was 176/24 then CD throws away 5/6 of all the information. It was only our lack of knowledge of hearing in the 1980's that allowed the marketing of "Perfect sound". Today we know better. Even admitted objectivists know better (provided their dog isn't barking).

I reiterate my point. Without knowledge of human hearing it is not possible to say whether or how well a given device captures an "audio" signal. All measurements are limited. Whether this matters depends on how the measurements are to be used.

 

[amirm]

Seeing a comment from the other thread and the continued talk about "measurements," I am starting to wonder if we are all in sync with what that terms means.

Here is a "transfor function" of your head and torso with respect to a sound source 30 degree to one side, e.g. your one speaker:

It shows how a signal with flat response gets transformed, i.e. changed, once received by your ear.

Is this a measurement?

Here is the sensitivity of the ear to each frequency, i.e. Fletcher-Munson equal loudness:

Each line shows equal perception of loudness while varying with frequency.

Is this a measurement?

 

[jkeny]

A lot of the measurement problem is making the right test signals. I wonder what is done in the way of impulse, modulated tones, broadband noise tests (ACPR-type tests), square waves, modulated square waves, and the like during design phases. So much that we could measure (or are measuring) is not reported and correlated to listening. Frequency response, basic THD/SNR/SINAD, and power output is a pitiful small sampling of the measurement suite we can and should be doing. Some of that may be because there is no real incentive to do so, and some of it may be because they are not standardized. And all the usual reasons (e.g. nobody wants to be first to report a new spec; the competitors will then provide a slightly better number, natch).

Don, I don't agree. If we take the example of hearing aid devices where there is a great finacial incentive to get it right & do the necessary measurements - are you saying that they just don't do the tests/measurements which would fully characterise their hearing aid & this is why it isn't perfect? I would expect such manufacturers would first develop a full size device that perceptually had no issues for users & then proceed to the difficulty of trying to miniaturise it to a suitable level. But I don't see this happening or am I wrong?

 

[Speedskater]

That's interesting. By "demonstrate two phono cables that have different soundstage's" do you mean through measurements or through listening or through both? How is the demonstration conducted?
............................................................

The listener picks the cables, the musical source, the associated equipment, the volume and the amount of time needed to listen to cables 'A' and 'B'.
Long before starting the test, the listener finds selects 3 cables A, B & C. With A & B having very different soundstage's and C having a soundstage similar to A or B.
With just measurements, it would be easy to say whether C's soundstage is similar to A or B.

Why or how? Because all accurate cables have about the same measurements, if the cables soundstage's are different, then at least one of the cables will have unusual measurements.

 

[Speedskater]

CD itself is a lossy encoding method. It throws away information that is in the original sound (if analog) or high resolution master. If the master was 176/24 then CD throws away 5/6 of all the information. It was only our lack of knowledge of hearing in the 1980's that allowed the marketing of "Perfect sound". Today we know better. Even admitted objectivists know better (provided their dog isn't barking).
.............................................

This begs the question:
What is 'information' ?

To me the 'information' is what a listener in a hall listening to a live un-amplified musical performance would hear. That listener would be positioned at the best seat in the hall.

 

[jkeny]

Sorry, Amir, I don't get your point. Are you saying that subjective testing is also a measurement? Sure, I don't think there is any disagreement about this.
What is being asked is - how well suited are the test signals used to establish that these are the actual lower thresholds of what's audible?

I don't just pull this out of the blue:

- I remember Martin Mallinson making the claim on his RMAF video that some people (in ESS) can evidently hear features well below the noise floor which were previously held by the majority of engineers to be inaudible. This was said to be confirmed in blind testing.
Large changes in the values of variables in the state space are said to contribute to non-periodic steady-state noise normally invisible using conventional measurement techniques. This is visible when instrumentation is used to examine the Noise-vs-DC Offset.
This is said to be minimized in ESS products by techniques designed to control the state-space related noise by rapidly quenching state variable excursions
I don't think the video is available anymore​

- Rob Watts in his comments about the new Chord Dac has claimed that he simulated -200dB noise floor modulation injection into the digital signal & it's presence is audible - see this video at 21:20
[video]https://youtu.be/10k_yMHCncY[/video]​

You might consider that these are commercial products & therefore dubious claims but these well known engineers are willing to make these statements & risk their reputation!

 

[amirm]

Sorry, Amir, I don't get your point.

I am not saying anything yet . Just asking a question for now.

What is being asked is - how well suited are the test signals used to establish that these are the actual lower thresholds of what's audible?

No, I am just at terminology state right now. I like to hear from people who are opposed to measurements, what they think the term means.

 

[Ron Resnick]

There is plenty of room for discussions about both objective measurements and data and how they effect audio system performance and also subjective listening impressions and how our ears tell us things that audio science can not yet explain.

Does this highlighted sentence, in my original quote above, seem confrontational, controversial and "(hugely) inflammatory to the other camp"?

Peter, I think your highlighted sentence is fair, balanced and perfectly accurate.

 

[esldude]

Here you bring up a sore point. Engineers are so cock-sure about their methods, like scientists would never be. Scientists know that, the more we know, the more we know what we don't know.

I think there is a bit of misunderstanding here. Most engineers are not so cock sure in the sense they don't know of areas that aren't a certainty. Now yes, if your job is to engineer a solution, you don't get solutions based upon what is not known. You have certain procedures worked out to reliably give performance within known parameters. Now when those areas they work in and have experience with are involved they can be rather sure. If you suddenly told a bridge designer there were flaws in bridge design indicating they don't really work, you probably aren't going to get far interesting that designer. He has built bridges and knows the relevant ins and outs. Can point to bridges that indeed do work to convey loads within design limits without falling down. If he is tasked with something near the edges of the possible or the known, there may be special research or testing to validate a design.

 

[esldude]

Don, I don't agree. If we take the example of hearing aid devices where there is a great finacial incentive to get it right & do the necessary measurements - are you saying that they just don't do the tests/measurements which would fully characterise their hearing aid & this is why it isn't perfect? I would expect such manufacturers would first develop a full size device that perceptually had no issues for users & then proceed to the difficulty of trying to miniaturise it to a suitable level. But I don't see this happening or am I wrong?

Do a Google Scholar search on almost any topic you think relevant to high end audio design. You in almost every case will turn up research and testing related papers regarding the design of hearing aids. Usuallyl dozens and dozens of them. These aren't your usual tests. There is considerable work about hearing intelligibility issues. Many innovative methods of investigating such. Or at least innovative vs your usual high commercial white paper pulp.

 

[jkeny]

Do a Google Scholar search on almost any topic you think relevant to high end audio design. You in almost every case will turn up research and testing related papers regarding the design of hearing aids. Usuallyl dozens and dozens of them. These aren't your usual tests. There is considerable work about hearing intelligibility issues. Many innovative methods of investigating such. Or at least innovative vs your usual high commercial white paper pulp.

Sure but I think you missed my point - yes, hearing aid designers DO use lots & lots of measurements that aren't used in the usual suite of audio equipment tests. They are motivated by financial rewards in the marketplace to get it right i.e a far bigger market than music playback systems. So, they have the money & the motivation to do any measurements yet none of them have developed a hearing device that is without issues. Why is this the case if we can measure anything - surely in the decades of research into hearing aids the full characterisation of a device for the ear to aid hearing would have been developed? Surely all the possible measurements are available to these developers but yet no hearing aid has been developed that hasn't got issues.

So my question is - do they have the measurements for the perfect hearing aid model but they are unable to build it

Or, have they not been unable to measure everything needed to specify the perfect hearing aid?

My bet is on the second option

 

[PeterA]

Peter, I think your highlighted sentence is fair, balanced and perfectly accurate.

Thank you Ron for answering one of the core questions I asked in the OP. I appreciate it.

No one has yet written me, publicly or privately, agreeing with Amir that this statement is confrontational, controversial or "hugely inflammatory to the other (objectivist) camp." If some think it is, I would really like to know so that I may avoid offending anyone in the future.

 

[DonH50]

Don, I don't agree. If we take the example of hearing aid devices where there is a great finacial incentive to get it right & do the necessary measurements - are you saying that they just don't do the tests/measurements which would fully characterise their hearing aid & this is why it isn't perfect?

Not at all. I suspect much measurement is done during design and simply not reported, whether because it is considered proprietary or because it provides no market advantage I could not say.

I would expect such manufacturers would first develop a full size device that perceptually had no issues for users & then proceed to the difficulty of trying to miniaturise it to a suitable level. But I don't see this happening or am I wrong?

I have insufficient experience with hearing aides to debate their design and achievements/shortfalls. I don't know what their manufacturers do. I have some experience with audio designs and designers but long enough ago that I would not dare compare it to today's designers.

Aside: I do think referring to engineers as a group as "cocksure" is a disservice to the profession, but of course I am biased. And, after spending much of career on essentially applied R&D, I have met plenty of scientists who are just as assured of their results and are proven wrong...

Time for me to take a break. - Don