Preference vs. "correct" and Accuracy

[jj]

There's something interesting that happens in audio systems, in particular, many people have found, some of them under analytical conditions, that they prefer signals with some moderate-level, low-order distortions to a signal that is analytically accurate.

Arguments rage endlessly on audiophile forums between those who believe that the distorted signals (and they are distorted, measurements don't lie, don't cheat, and don't use perception) are "more accurate" and who generally reject the measurements as part of an appeal to ignorance, vs. those who argue that the "right signal" is the analytically accurate signal.

A few people make a third argument, that the measurements are right, but that the distorted signal may be PREFERRED. This isn't a very popular point of view, despite the simple facts of the matter, in particular that it is easy to show that some kinds of distortion increase a sensation of dynamic range, some other distortions (as well as some kinds of linear filtering) create a sense of wider or more enveloping sound from a 2-channel system, and that many of the distortions raised by the "it sounds more accurate therefore it is more accurate" crowd are flatly below the threshold of audibility, in fact often well below (not just a bit) the atmospheric noise level at the eardrum.

There is a basic philosophical question here, one that usually gets obscured in all of the ranting, raving, and accusations of scientific misconduct, etc, and that is the question of which IS right.

Are any of them right?

As one might guess, if one knows what I do for a living, I have opinions.

What I wonder is what others think.

There are some subsidiary questions, but perhaps I'll leave them for now, and raise the issues later on if they arise.

 

[Soapy Sam]

Hey, jj!

I told you once that I prefer MP3 to WAV files. Thiere may be any number of technical reasons, which you know far more of than I.
In nature, all sound is distorted.I mean what is a footstep in a dark room at night, in the rain supposed to sound like?

Each of us carries a sophisticated sound processor between our ears. The brain gets used to unscrambling distortions. We learn them- especially in specific locations. (Think a mechanic, listening to an engine).

In the case of music, most of us these days hear most music in recorded format. That's what we're used to. I once heard an opera singer perform "Both Sides Now"- she was an incomparably "better" singer than Joni Mitchell. She was crap. The song didn't sound like I expected it to and my filters didn't handle it well.
If I heard her sing it fifty times, the filters would learn.

I've been disappointed by live performances of orchestral music, because it didn't sound like I expected it to.
On CD, I have 4 versions of Smetana's "Ma Vlast". I like the oldest one best and the other three in order of purchase and of play frequency. Coincidence?

Recording several versions in different formats, then shuffling them is interesting.
Played at random I still find it easy to say which I like best, though I may be unable to say which version is which.

Our ears learn though. Preferences do change with time- but maybe by "time" I mean "exposure". We learn to like what we know.

 

[Aepervius]

A distorted signal cannot be more accurate. It *might* be preferred because of the audio-taste of the person, but it is NOT more accurate. An accurate representation of a signal is a identical undistorted output of the original signal.

The rest (whether with or without distortion is better) is only a question of taste.

PS: anyway due to various factor, one being the loudness war, the other the equipment used to record/play the audio signal, you never get an accurate reproduction of a signal. But in most case (except loudness war) the reproduction is good enough that our ears won't hear a difference (10-22khz range).

 

[Soapy Sam]

What if the ability to recognise (and remove) the diistortion is built into the receiver?

For example, the perceived sound of a gunshot carries information about the distance from the gun. That information is not coded anywhere in the signal.

I guess what I'm thinking is this- the audio industry in the main reproduces music, but does so to feed information to an acoustic recognition system evolved to unscramble severely distorted sound. Maybe we actually derive satisfaction by unscrambling distortion, so a "pure" sound is ...rather dull?

 

[jj]

A distorted signal cannot be more accurate. It *might* be preferred because of the audio-taste of the person, but it is NOT more accurate. An accurate representation of a signal is a identical undistorted output of the original signal.

There are some interesting issues in this regard. When we record a signal into two channels, we lose 99.95% or more of the information that is actually present in any good acoustic venue (i.e. we do not know the actual spatial distribution of energy in a room, something we learn to sample with our two ears as a baby, assuming we have two functioning ears that can even be substantially impaired).

Some forms of distortion can somewhat mimic the perceptual effects of that, to some small extent. (in particular differentially different distortion in M vs. S in an LP)

So, the original pressure or linear combination of pressure/velocity at two points may be more "accurate" for those points, but what is "accurate" to the original venue, assuming for the minute there was one that was recorded in the classic style, say with Blumlein?

Tougher question than it looks, btw.

 

[jj]

What if the ability to recognise (and remove) the diistortion is built into the receiver?

For example, the perceived sound of a gunshot carries information about the distance from the gun. That information is not coded anywhere in the signal.

Now a gun shot is a particularly interesting issue here. A gun shot, at its source, is well over the level at which air is a nonlinear medium. As a result, the spectrum (and waveshape, of course, although "waveshape" and 'nonlinear' are an interesting pair indeed) will change by itself, even without any reverberation involved, until the gunshot gets down into the 100dB peak SPL range or so. (air actually distorts at all levels, but the distortion goes up hugely with level, over 140dB air just isn't linear. at 120 it's easily measured to be nonlineary. at 90dB one can measure it, carefully with good equipment)

So even in a free field (nothing but air) you'd hear a gunshot change with distance.

In addition to that, we have reverberation in the usual case, and one of the cues that we use for distance estimation (and it is estimation only) is the ratio of direct (first arrival, which is, barring occlusion, always there first) to reverberant sound arrival.

There is a basic property of any acoustics space, called the "critical distance", which is the distance from the source that the reverberant energy is equal to the direct energy. Closer, more direct, farther, more reverberant.

So that also comes into play in a gunshot scene.

Gunshots are quite interesting in that they clearly demonstrate the nonlinearity of the ear, though.

 

[Soapy Sam]

jj- it never occurred to me before that the "audio" industry is essentially a music (and speech) reproduction industry and does not obviously pay much attention to recording or reproducing other sound.
In a movie someone points a gun and pulls the trigger. We hear a bang. But was the bang actually made by the gun, or was it added afterwards?
Do we record and reproduce sound for any reasons other than speech or music?

 

[Paulhoff]

Do we record and reproduce sound for any reasons other than speech or music?

Whale songs, bird songs, sounds of rain, ocean, need I list more.

Paul

 

[Bodhi Dharma Zen]

Audiophiles and their discussions about accuracy... I love classical music and I have a fairly good stereo and multichannel equipments. I did my homework studying such discussions in several audio forums.

My conclusion, most audiophiles are really audiophools, they shout repeating like parrots concepts that sound appealing.

"Less distortion = better"

"The goal of an audio chain is accuracy"

Yeah, sounds logical at a first approach. But when you dig in to what exactly is distortion, and how can you measure accuracy you will find that both are theoretical concepts used by the industry just to induce you to buy their stuff.

Two easily understandable things.

1) There is nothing like a sound that does not interact with the surroundings.

2) A piano will sound different in two different locations.

So, right from the beginning, you can't expect to have a sound that it is not distorted (by interacting with the environment) and it is idiotic to assume that a piano "sounds more like a piano (read more accurate) in one location than in another.

Attempting to discuss a particular set of distortions, THD for example, lead us to an empty place because human ear is unable to differentiate between, say, 0.0000001 and 0.0001, so both figures are irrelevant.

Anyway, I'm done discussing audiophooly, all I do now is to enjoy my collection of MP3s using cheap cable in my expensive equipment (by the numbers, it can put 115dB from 18 to 17Khz), and yes, it sounds even better than equipment costing about 20 times more.

 

[Mercutio]

I remember an article in Consumer Reports, on potato chips. The CR people were testing some dozen or more different brands of potato chips against one another, using trained tasters, as they often do. (I don't know if they also surveyed non-trained; sometimes they do that, but not always.) What struck me, in the article, was a comment that American potato chip eaters (not the experts, but the general public) actually preferred a somewhat stale-tasting potato chip over a fresh one, presumably because that is what they were accustomed to.

Completely unrelated, I also recall two separate radio programs in which expert and novice judges were rating the sound of a new design of carillon bells (in one case) and a new design of violins (in the other--actually, the whole family of violin instruments); in both cases, the harmonics of the new instrument were vastly different from the harmonics of the traditional forms, and in both cases (more decidedly with the bells) the untrained judges preferred the new designs over the old, while the trained judges preferred the old over the new.

So... can we prefer, due to experience, something that is "distorted" over something that is not? Clearly, yes. Does that make it "right"? I could easily argue for at least options 2 and 3 as both being "right", for given values of right.

What makes the "preferred distortion" any different from, say, adjusting your equalizer to tune your stereo to your room? If I know, for instance, that I have hearing loss over part of the spectrum, is a "distorted" equalization literally more accurate, once the human element is accounted for? And do we draw a line, then, between distortions that compensate for hearing loss, say, and distortions that compensate for some unspecified "preference"?

Once again, a jj post has intrigued me... no surprise.

 

[Paulhoff]

I remember so well arguing with my vinyl loving tube friend over a record of a piano being played over his system some many years ago. He said, that is how a piano should sound, and me being someone who likes to think that he thinks said to him, how do you know that is how that piano should sound. You never heard that piano that you are listening to now, you never heard it in the room that was played in, you never heard the person playing the piano and there are many more things that I haven’t thought of that come into play with the sound of that piano at the time it was recorded. So I said to him, not knowing all that, how can you tell me that is how a piano should sound.

Paul

 

[JoeEllison]

A distorted signal cannot be more accurate. It *might* be preferred because of the audio-taste of the person, but it is NOT more accurate. An accurate representation of a signal is a identical undistorted output of the original signal.

Maybe it can be, for a certain definition of "accurate". There is no such thing as a "identical undistorted output of the original signal" in part because any recording of a signal involves a certain amount of compromise of the original. I can imagine simple ways that "distortions" to the playback signal could better replicate the original listening experience better than an "undistorted" version.

 

[roger]

What do you mean by accuracy?

Here's what I'm getting at. A saxaphone, in real life, blats. It can rip your head off. It's visceral experience. Now, take a pair of electrostatic speakers, say that a classical musician might prefer, and a cheap pair of horn speakers. Probably the electrostatic speakers would measure more 'accurately', in terms of fidelity to the recorded signal. However, the person who lives for live jazz may very likely prefer the horn speakers because it recreates the experience of listening to a sax better, even if the frequency response is all off.

 

[jj]

jj- it never occurred to me before that the "audio" industry is essentially a music (and speech) reproduction industry and does not obviously pay much attention to recording or reproducing other sound.
In a movie someone points a gun and pulls the trigger. We hear a bang. But was the bang actually made by the gun, or was it added afterwards?
Do we record and reproduce sound for any reasons other than speech or music?

Movies are completely sound-tracked after the fact, in general.

Movie sets are really loud and annoying.

So do a google on "Foley" (no, NOT catheters)

 

[jj]

Maybe it can be, for a certain definition of "accurate". There is no such thing as a "identical undistorted output of the original signal" in part because any recording of a signal involves a certain amount of compromise of the original. I can imagine simple ways that "distortions" to the playback signal could better replicate the original listening experience better than an "undistorted" version.

This is the tip of the iceberg.

When you reduce a soundfield to 2 (or 5.1 or 7.1) channels, you are throwing away just about all of the actual information in the room.

So, what we get into the transmission device (file, LP, CD, tape, radio signal) is heavily lossy.

I won't use the word "distorted" because I try to reserve that (although many people do not) for lossy, nonlinear processes.

But there is a huge amount of information lost at capture time.

Consider. At one point in space, it takes 4 variables to express a sound wave passing through that space. (not just one as most people assume) All 4 of these variables represent energy in the soundfield.

If we are to capture a highly diffuse soundfield accurately, let's see now, that's what? Oh, spatial sampling every half-inch or so, 4 variables at each.

Obviously, nobody does this (well, sonar devices do, but we don't have the government budget to play with here) in consumer devices.

What can we capture? We can capture an illusion of what there was. It might be possible to compare the illusion, using a human listener, to the original. Good luck with that now. Remember, it has to be a blind test, and the stimulii have to be close enough that the listener can't ID the signals and break the blinding.

Note, now, I'm even assuming that there WAS an original acoustic venue. For many (most) modern productions it's synthetic, from stem to stern, port to starboard. For those, what is "accurate" in the first place, now?

jj

 

[dudalb]

Wow I amazed.
A discussion on Sound Equipment and extremeskeptic has not shown up.

 

[R.Mackey]

A few people make a third argument, that the measurements are right, but that the distorted signal may be PREFERRED. This isn't a very popular point of view, despite the simple facts of the matter, in particular that it is easy to show that some kinds of distortion increase a sensation of dynamic range, some other distortions (as well as some kinds of linear filtering) create a sense of wider or more enveloping sound from a 2-channel system, and that many of the distortions raised by the "it sounds more accurate therefore it is more accurate" crowd are flatly below the threshold of audibility, in fact often well below (not just a bit) the atmospheric noise level at the eardrum.

This is what I subscribe to, personally.

In my partly homebrewed and nontraditional music system, I have a single-ended tube line stage that I can use, or not, as the whim takes me. I've put it on the test bench and confirmed that, while it's pretty high quality, it basically introduces even-order harmonics and a soft landing when clipping, and that's it. It is not "more accurate." It is colored. Still, I find myself using it about 80+% of the time.

I am not even convinced that anyone even wants truly accurate reproduction, not really. I'm a regular at the LA Phil, and I play on my own a bit -- actually hearing real live music helps re-sync expectations. Compared to my favorite recordings, the real thing is sloppy in several ways (imaging, chamber reflections, especially background noise) although sometimes superior in others (dynamic range, physical impact).

My next system is going to be all high-efficiency, probably horns. Yeah, they're colored. And I don't care. I'll measure it anyway and won't make any apologies for its shortcomings, provided it does what I want. Maggie fans will probably hate it, and that's their right.

Sound is subjective. Measurements aren't. The two are not interchangeable.

 

[jj]

This is what I subscribe to, personally.

In my partly homebrewed and nontraditional music system, I have a single-ended tube line stage that I can use, or not, as the whim takes me. I've put it on the test bench and confirmed that, while it's pretty high quality, it basically introduces even-order harmonics and a soft landing when clipping, and that's it. It is not "more accurate." It is colored. Still, I find myself using it about 80+% of the time.
...

Sound is subjective. Measurements aren't. The two are not interchangeable.

As it happens, while I have moderately different preferences, I agree exactly that it's preference, that there is no "exact" to hew to in any substantial fashion.

For that position, I've been reviled by Arny Kreuger, Howard Ferstler, Michael Fremer, and some other high-end guy who's name I've forgotten, within 24 hours.

(for the uninitiated, that's two hardcore "objectivists" and two rabid "subjectivists")

JJ also looks at your name, and then across the street into Woodinville, and sees an "LOUD" sign where another name used to be, and wonders.

 

[Paulhoff]

Sound is subjective. Measurements aren't. The two are not interchangeable.

That is not true about measurements, I see measurements used subjectively all the time.

Paul

 

[Soapy Sam]

All human activity has some subjective element, surely?