"Things that cannot be measured"

[j_j]

I still think an assessment with ear and brain and a defined reference is superior in terms of measurement, in terms of what matters to the listener.

In other words, an ABC/hr double-blind test.

 

[j_j]

Reproduced music is all but live music. Best you can do is maintain tonal accuracy through the producing chain and illusion of orchestra. Don't even dream of replicating the acoustic power of a band in your room

Capturing the actual soundfield in the original venue? Well, yes, you might be able to do that with 10,000 channels.

So, yeah, that's kind of a non-starter. There's also the issue of such a recording apparatus actually affecting the original venue's soundfield, same problem on rendering in the listening space.

 

[j_j]

Power of an orchestra might be 70W, but in what [concert hall] volume? At what listening range? You do not need that power (level) in your room, at a few meters (It will deafen you). And even it a speaker efficiency is 3% (sounds way low, maybe another ‘matter of definition’ disconnect), yet regardless, there is hardly a problem providing 2.3kW of (now electrical) amplification.

This emphasis on power is simply not the issue. The real issue is the complexity of the soundfield in a listening venue.

So can we stop this sub-thread?

 

[j_j]

Ooookaaaay... I have reached my limit with this thread, with the insults about "put your blind test" that demonstrate appalling, regressive, destructive anti-scientific insanity, etc.

LISTEN UP EVERYONE.

In an original venue, when one exists (and for most recordings nowadays that is not the case), the instruments, performers, etc, create a soundfield in hall. A very, very complex soundfield. The acoustics of the room create some of this complexity in the room. The air movement in the room, the result of warm people in a cooler room, ALL CONTRIBUTE TO THE SOUNDFIELD IN THE ROOM THAT SURROUNDS YOU AT ANY GIVEN LISTENING POINT.

This soundfield is very, very, complicated, time varying, and varies substantially over short distances.

YOUR HEAD MOVES IN A SOUNDFIELD

Yes, yes, it does that. You do that. Anyone who has ever bothered to watch people in a listening venue will see that. When you move your head, you are optimizing FOR YOUR OWN PERSONAL VALUES, a listening position FOR YOU inside that space.

What is the complexity of that soundfield? Well, let's say, in order to capture it around your head, you could probably do so with a geodesic array of pressure microphones on a 12mm spacing in a 1 meter radius around your head. Yeah, so how many channels is that? Answer: Too many to even think about unless you're building the sensor array for a giant sonar system that costs 3 billion dollars or something like that, but wait, there's even MORE problem. Each sensor takes some energy out of the soundfield, so guess what, boys and women, YOU JUST CHANGED THE SOUNDFIELD BY OBSERVING IT.

So, that's a reality you just have to live with. And remember, just for grins, that 'time-varying' part that comes from air currents, air handling devices, temperature gradients, etc, that are by their very nature unpredictable, except in the aggregate.

Now, what does the "hi-fi" industry do? First, they capture this in a limited number of channels, from 1 to maybe 48 or so, and then they MIX THAT DOWN INTO 2 channels. So, you've already eliminated damn near all of the information in the venue. Congratulations, you'll never actually have this in your home. Nope. Not now, not ever, never, because you can't get around that "affecting the soundfield" issue even if you were richer than the US government building a state of the art attack submarine. (Yes, that's an applicable comparison, figure it out for yourself.)

So, now, what do we have, two channels of information that is missing most of the original information. But now, we need to back a minute and ask "how much of that information can the ears actually detect, even considering motion allowing one to sample the space". The answer is, happily, that the ear can not even possibly detect almost all of that information. So, fortunately, we are not completely at a loss here.

But, remember that motion thing. So, ok, maybe we can use the idea of an "artificial head" recording (binaural recording). Well, great, works ok into headphones, but have you ever considered just how hard it will be (i.e. impossible) to handle that head movement thing? Still, it's an option for headphones, and it can work well, for limited applications. But is it accurate? Oh (*&(*& no, not only is most of the information missing, there is frequency shaping from the head, etc, AND THAT IS HOW IT WORKS. So, oh those who breathe accuracy, HOW DO WE MEASURE THE ACCURACY HERE? Lest the subjectivists now start to cheer, sorry, bzzzt, your head isn't the same as the artificial head, so sorry, the perceptual outcome isn't accurate, either.

What do we do with loudspeakers? Well, we put two signals into two loudspeakers. The direct radiation from the two loudspeakers reaches your ears first (unless we're talking about a few very odd speaker systems, that we shouldn't be even considering here for "realistic" reproduction). It is soon accompanied by room reflections, reverberation, etc, that is added to the original captured (or synthesized) signal. But we can still hear direction, and get a sense of the original venue, so how does that work?

Well, in a nutshell, first arrival is very important under 500Hz, and first arrival of the signal envelope likewise over 2kHz. So, the interaction of two speakers, the soundfield in the ROOM, and the signal captured in the recording, all interact to create both a time and intensity tradeoff that sounds kind of maybe sorta a bit like the original.

And that's ***ALL*** you get from 2 channels.

It's a fact, live with it.

 

[BluesDaddy]

We can take it - the concept - even further. At some distant future, thanks to progress of the AI, everyone will have a personal copy of “pocket Genesis” (Peter Gabriel‘s Genesis that is), “pocket Pink Floyd” (don’t get me started, whose Pink Floyd), “pocket Beatles” (maybe we can settle on just George)... Upon demand, we would summon those to play a tune-or-two for us. And it will be undeniably, unquestionably “true to the source” high-fidelity experience... Someday.

Now you'll have to define what you mean by "source" and "true to". Since none of those recording "events" could possibly have the detailed information available to create or recreate them, they'll have to be TONS of interpolation. That may be incredible, but no more "true to the source" than having tech that can get Jimi to play "Purple Haze" in my living room.

 

[Raindog123]

@j_j It's illuminating. Makes sense for the complexity of a [large] orchestra in a concert hall. Does the same applies to an all-electric rock concert? (Yes, there is/might still be a 'venue', but all the 'live' music goes through a limited number of speakers). What about a (relatively small) chamber or jazz trio/quartet? What about a solo singer? Can those be captured (eg binaurally) and represented with "high fidelity'?

What about 5.1? 7.1? Any hope for [better] high-fidelity there?

And, can you elaborate a bit on your "Well, in a nutshell, first arrival is very important under 500Hz, and first arrival of the signal envelope likewise over 2kHz"? You lost me there, quite a bit (even though I deal with multipath in my line of work all the time).

 

[scott wurcer]

And that's ***ALL*** you get from 2 channels.

It's a fact, live with it.

+10

 

[j_j]

@j_j It's illuminating. Makes sense for the complexity of a [large] orchestra in a concert hall. Does the same applies to an all-electric rock concert? (Yes, there is/might still be a 'venue', but all the 'live' music goes through a limited number of speakers). What about a (relatively small) chamber or jazz trio/quartet? What about a solo singer? Can those be captured (eg binaurally) and represented with "high fidelity'?

What about 5.1? 7.1? Any hope for [better] high-fidelity there?

And, can you elaborate a bit on your "Well, in a nutshell, first arrival is very important under 500Hz, and first arrival of the signal envelope likewise over 2kHz"? You lost me there, quite a bit...

7.0 ( .1 is faulty for other reasons) can provide a decent PERCEPTION of sound in the plane.

Even a solo singer, a-capella, has a complex sound field. Voices do not radiate in all directions, just like every other instrument, give or take maybe a very few. And even those that are almost omni tend to be dipoles or bipoles. So the capture, unless it's electronic end to end (no acoustics involved until loudspeakers) is missing something every time.

Pure electronica, well, it's art, and it's artificial to begin with, so it is what it is. What it's not is a sensation one would get in a natural or even synthetic venue, unless such has been added.

 

[j_j]

*

It's simply not correct, two microphones at a distance can do it, even with a symphony orchestra,

Let me ask you. I take one point in a room. How many channels are required to capture the sound field at that point in the room? Be specific, and tell me how many channels of data exist at that one (any one) point.

Now, you can apologize for the vile professional accusation you've made, and retract it. Furthermore, you shall agree to take responsibility for that vile disparagement. I'm tired of listening to your ignorant blunders about blind tests, and now you act like you actually know something about acoustics, and you've proven that you don't.

You are clearly ignorant of the most basic aspects of acoustics if you think that two microphones, each feeding one channel into a recording, of any sort capture the entire details of the soundfield even around one point, let alone a listener's head.

 

[Objectivist01]

1. Lie detectors suck. Pseudo science at best. So just a messy example.
2. Fair enough, but ultimately we should be able to (and I am confident we will be able to) model why particular emotional responses arise and predict them given a set of inputs (e.g., a room, a recording and measurements for a given audio device chain).
3. Facebook, google are farther along doing this sort of thing for consumer preference/emotional response than I think people realize. Things are progressing rapidly in this space. Prediction of individual thoughts based on group data plus some individual modeling is getting pretty good. I don't know if you count that as reading thoughts, but if the system makes the correct predictions for you, then it is essentially doing so (by some definition).

Are we sure about that? Is there some reading you can point me at? [I am genuinely curious about the state of it in the audio. Just a few posts above we were saying that any signal can be represented by a combination of amplitudes and phases... That and [linear] superimposition... ]

I think you need a bit more reading to be in Page of the people explaining stuff to you.

 

[escksu]

Hmm.... There are still many things we dont understand and dont know, esp. our brain. Maybe one day, we can really measure these emotions/feelings from our brain. Maybe one day, we are able to measure every aspect of sound.

 

[scott wurcer]

No I do not, it seems a bit like it is you who do not know how stereo works

Do we really have to put up with this?

 

[BluesDaddy]

Do we really have to put up with this?

Apparently not...

 

[AdamG]

Troll has been eradicated. We try to give lots of rope to members. Sometimes they fashion hammocks and hang out all chill, and sometimes they make nooses. Apparently the later was chosen.

 

[AdamG]

Do we really have to put up with this?

No Sir you do not.

 

[j_j]

Hmm.... There are still many things we dont understand and dont know, esp. our brain. Maybe one day, we can really measure these emotions/feelings from our brain. Maybe one day, we are able to measure every aspect of sound.

It would be wise to separate out the "sound" from the "human reaction to sound" in that sentence. One is extremely testable, the other is extremely plastic and subject to learning, mood, and all other outside influences.

 

[brbsnacks]

I believe what is being measured for headphones might only be half of what can be measured. The rest I dont think people have figured out yet. This is why I dont take the headphone reviews seriously at all. I'll take into account FR, noise, distortion, build quality, etc but I'll make my final judgement when I put them on and give them time. Especially the Harman curve which brought up more questions than answers when I read over the studies. I'm not trying to impress the masses of other people. I"m trying to impress myself.

 

[AdamG]

Just tossing this out. With all the talent and skill we have in this Community of all things Audio. Maybe someone will open a thread that focuses that collective intelligence on “Things that Can be Measured” instead of this (fill in blank) thread.

 

[CMOT]

Removing artefacts that look like noise is indeed common and been used for decades.
It is very, very different from attempting the impossible of adding "lost" information when there is no way to know what was lost.

Some of the de-noising approaches also add signal (e.g., when, for a given time increment, there is significant signal drop when noise is accounted for - essentially trying to reconstruct the signal that was masked by the noise). Isn't that adding lost information - based on the surrounding signal, the model makes a guess about what was lost.

 

[CMOT]

You cat train modern AI (eg, a neural net) on a bunch of faces, so the AI can compose a realistic-looking face following the trained pattern, or fill in missing pixels... But if I delete my mustache from a photo, it will never guess and reconstruct its shape or color. The same with audio - there can be (are?) some pattern-based [reconstruction] filters, and the AI now can even compose music pieces - again by identifying and following patterns... but it cannot recreate missing uncorrelated, non-redundant sound information.

The key issue here is "uncorrelated" - how sure are you when you delete the facial hair that all of the replacement pixels and the surrounding pixels aren't correlated to some degree. There are plenty of modern computer vision models that are generative and can do the equivalent of mustache replacement given scant evidence (but some evidence). Of course, if you remove something absolutely not correlated from the remaining data, no model can infer that the thing was removed. On the other hand, if the thing is something that can appear in that context (even just some smaller percentage of the time), then there are, again, many modern generative models that can do the equivalent of putting a mustache on your face and do a very credible, if not undetectable, job at putting it there. So reconstruction is possible by some definition of missing, but not by the absolute, entirely separable definition - in that case you would have to appeal to top-down knowledge.