Our perception of audio

[andreasmaaan]

PS: Did you refer to your post #296

Sorry, I meant post #289.

Will give that article a read, thx.

 

[Cosmik]

So you disagree, then, with Watkinson’s list:

https://www.thebroadcastbridge.com/...echnology-part-15-a-catalogue-of-shortcomings

I like the cut of that bloke's jib .

 

[andreasmaaan]

@svart-hvitt i read the article. Predictably, perhaps, I disagree with some of his views, and my reasons are no doubt well-documented elsewhere on the forum Ofc I agree with a lot of his views too though.

 

[Snarfie]

My impression regarding our perception of audio is since a few months (I'm 59) that a lot of people really don't have an idea how something must sound. Thats probably because most of us (compared with a proper acoustic/dsp treated studio) have a lousy acoustic listening room. The differences that i heard between cables DAC's etc for many years are obsolete compared with the sound that i have now thanks to room correction (RC).

A piano like a Bosendorfer sound quite different like a Steinway but if you have a room like me see picture below (grey line) the two piano's almost sound more or less the same due to lots of resonance that cuts/dominate mid a lower frequencies. For instance Ahmad Jamal who played the Steinway sounds with RC way better than without (see white line). Basicly the Steinway (with roomcorrection) does not sound anymore like a snare drum in the higher frequencies what a relief was that

But if i use the RC than the sound is more or less as it should be. No it is not spectacular but from one on the other moment the band is playing like a team/band in balance. Yes i did compare after RC cables dac's etc but the balance always is there the differences as before where not huge. It is not so that i hear now much more details with RC but i hear much more the intimate interaction an details between band members that I did not hear in 50 years thanks to modern technology the fun is back.

So before trying to hear differences between gear I would suggest first use RC (get your accoustics as much in order as possible) an if necessary than try to tweak your gear.

ps. above opinion is strictly my experience an the specific acoustics of my listening room can't speak for any other listening room.

 

[Krunok]

Correcting speakers response in your room is IMHO the best thing you can do to improve SQ, second only to buying better speakers. And even with them you'll need to do room correcion as well.

 

[Floyd Toole]

@Floyd Toole , your position, and many ASR members’, on point source seems to be:

1) Point source is ideal
2) Point source is a theoretical concept that cannot be fully implemented in real life
3) Psychoacoustically, traditionally stacked drivers work just fine, as evidenced by listening tests where researchers used traditionally stacked driver speakers

You didn’t really write (1), but I have never met anyone who hasn’t been in agreement that point source is ideal. So I put (1) there to make this discussion a target for a wider audience.

On (2): Because something, an ideal, is impossible to work out 100 percent in reality, does it mean we should give up the idea of the ideal? We have different ways of engineering things:

a) You can tweak and perfect an old proven design
b) You can come up with a better idea, still unproven, and set out on a journey to realize that idea even if you know that the idea is a theoretical one that can never be realized 100 percent.

I believe I am a (b) person; if someone tells me my idea is an impossibility, I may still spend some time on it because I know option (a) is unchallinging and boring. And I believe that if I realize just a part of a BIG idea, I may still create something which is greater than a perfect realization of the conventional, yet smaller idea.

On (3): All (?) psychoacoustic studies that we normally refer to have been done on conventionally stacked design. How can one use these studies in arguments on the pros et cons of point sources? Isn’t it like running in circles? Shouldn’t point source (attempting) designs be used to evaluate the value and validity of point source design (attempts). My point is, how strong is the psychoacoustic research that we normally refer to? Is it so strong that we can disregard attempts - existing and future - to realize the point source ideal?

Existing point source designs have been proven to have more robust directivity than even a two-way design. So I cannot understand that @Floyd Toole and others think chasing the point source ideal is futile («Good luck with that!»).

Because existing point source designs already have superior directivity off and on axis, I think one should spend more time discussing the drawbacks (and positive sides) of these point source designs - as evidenced by measurements and data - instead of attacking attempts at bridging the gap between idea and real-life engineering.

Let’s not kill the idea of point source without data to support the position that point source is futility.

Some thoughts:
1. I know of no real-life vocal or musical sound source that behaves like a point source. In fact, as shown in Figure 10.18 human voices and many musical instruments have directivities not unlike conventional forward firing loudspeakers. Humans, therefore, are well adapted to listening to the direct and reflected sound fields of such sources - not point sources. Approximations of point-sources are used to measure concert hall reverberation times (pistols, firecrackers, popping balloons, crudely "omni" loudspeakers) on the assumption that an entire orchestra will at different times radiate some sounds in all possible directions . . . or not. It is another of those "academic" notions that becomes a practical norm but which is not an accurate representation of what happens in life. At an Acoustical Society meeting a few years ago there were a few papers by academics who were measuring the omnidirectIonality of popping balloons. To me that is an indication that they were running out of thesis topics that had any potential to contributing to the useful body of knowledge.
2. As discussed, with measured evidence, in Chapter 5, the steady-state room curve at frequencies above about 500 Hz is well estimated from early reflections; first reflections from floor, ceiling and walls. This means that most of the physical energy is in those components, and that perceptions will be dominated by them. Sounds radiated in other directions encounter multiple surfaces and much longer propagation paths en route to ears and are substantially attenuated. Simultaneous and temporal masking will further reduce the audible contributions of those later sounds. As I have said, I know of no evidence that indicates an inherent superiority of "perfect" omnidirectIonality, even in horizontal and vertical planes.
3. Usually, and I assume in this discussion thread, the relative virtues of loudspeakers are evaluated in two-channel stereo. Adding the many reflections from omni loudspeakers is known to reduce the audibly stark illusion presented by hard panned L & R sounds - softening the images and reducing the impression of sound emerging from a monophonic point in space. Also, inevitably, the clarity of the "double-mono" phantom images, including the star performer in the center, is reduced. Some like this, some don't. But all of it exists because of the directionally and spatially deprived performance of stereo - which sadly we are stuck with as the musical norm. Some of us have moved on to multichannel enhancements of stereo, in which one can exercise some control over "reflections". Because timbre and localization are dominated by the direct sound, omnidirectional loudspeakers would not be advantageous.
4. Finally, until mixing and mastering engineers start using omni loudspeakers in somewhat reflective rooms the "circle of confusion" is aggravated when customers start editorializing on things at home. The art cannot be preserved. It is like buying an old master and illuminating it with colored light. Personal preference prevails, but at least the light can be turned off for others.

If I were to be able to return to the world of audio research I would work towards a better understanding of multichannel upmixing. Dr. David Griesinger, when at Lexicon did a superb job of stereo enhancement in the Logic 7 upmixer. It pretty much left the stereo soundstage alone while adding tasteful amounts of envelopment. In the best implementation it was adjustable. Sadly that is gone. Now I am experiencing Auro3D upmixing in an SDP-75 and at its best it is quite satisfying. It too is adjustable. It quickly becomes clear that no upmixing algorithm setting is satisfying for all program material. However, it also becomes clear that reverting to raw stereo is frequently a backwards step. The same is true of loudspeaker directivity patterns. Fortunately, humans are very adaptable.

 

[svart-hvitt]

Some thoughts:
1. I know of no real-life vocal or musical sound source that behaves like a point source. In fact, as shown in Figure 10.18 human voices and many musical instruments have directivities not unlike conventional forward firing loudspeakers. Humans, therefore, are well adapted to listening to the direct and reflected sound fields of such sources - not point sources. Approximations of point-sources are used to measure concert hall reverberation times (pistols, firecrackers, popping balloons, crudely "omni" loudspeakers) on the assumption that an entire orchestra will at different times radiate some sounds in all possible directions . . . or not. It is another of those "academic" notions that becomes a practical norm but which is not an accurate representation of what happens in life. At an Acoustical Society meeting a few years ago there were a few papers by academics who were measuring the omnidirectIonality of popping balloons. To me that is an indication that they were running out of thesis topics that had any potential to contributing to the useful body of knowledge.
2. As discussed, with measured evidence, in Chapter 5, the steady-state room curve at frequencies above about 500 Hz is well estimated from early reflections; first reflections from floor, ceiling and walls. This means that most of the physical energy is in those components, and that perceptions will be dominated by them. Sounds radiated in other directions encounter multiple surfaces and much longer propagation paths en route to ears and are substantially attenuated. Simultaneous and temporal masking will further reduce the audible contributions of those later sounds. As I have said, I know of no evidence that indicates an inherent superiority of "perfect" omnidirectIonality, even in horizontal and vertical planes.
3. Usually, and I assume in this discussion thread, the relative virtues of loudspeakers are evaluated in two-channel stereo. Adding the many reflections from omni loudspeakers is known to reduce the audibly stark illusion presented by hard panned L & R sounds - softening the images and reducing the impression of sound emerging from a monophonic point in space. Also, inevitably, the clarity of the "double-mono" phantom images, including the star performer in the center, is reduced. Some like this, some don't. But all of it exists because of the directionally and spatially deprived performance of stereo - which sadly we are stuck with as the musical norm. Some of us have moved on to multichannel enhancements of stereo, in which one can exercise some control over "reflections". Because timbre and localization are dominated by the direct sound, omnidirectional loudspeakers would not be advantageous.
4. Finally, until mixing and mastering engineers start using omni loudspeakers in somewhat reflective rooms the "circle of confusion" is aggravated when customers start editorializing on things at home. The art cannot be preserved. It is like buying an old master and illuminating it with colored light. Personal preference prevails, but at least the light can be turned off for others.

If I were to be able to return to the world of audio research I would work towards a better understanding of multichannel upmixing. Dr. David Griesinger, when at Lexicon did a superb job of stereo enhancement in the Logic 7 upmixer. It pretty much left the stereo soundstage alone while adding tasteful amounts of envelopment. In the best implementation it was adjustable. Sadly that is gone. Now I am experiencing Auro3D upmixing in an SDP-75 and at its best it is quite satisfying. It too is adjustable. It quickly becomes clear that no upmixing algorithm setting is satisfying for all program material. However, it also becomes clear that reverting to raw stereo is frequently a backwards step. The same is true of loudspeaker directivity patterns. Fortunately, humans are very adaptable.

Thanks, @Floyd Toole !

AES Fellow John Watkinson writes here about the aperture effect:

https://www.thebroadcastbridge.com/...speaker-technology-part-5-the-aperture-effect

It seems like you think this concept is of little value in audio?

Could you point out where Watkinson’s reasoning is flawed?

 

[Floyd Toole]

Thanks, @Floyd Toole !

AES Fellow John Watkinson writes here about the aperture effect:

https://www.thebroadcastbridge.com/...speaker-technology-part-5-the-aperture-effect

It seems like you think this concept is of little value in audio?

Could you point out where Watkinson’s reasoning is flawed?

It is a real effect and, as he points out, in loudspeakers it shows up in off-axis measurements, which is why such measurements need to be made. It cannot be avoided in practical loudspeakers - nor in the musical instruments generating the sounds we try to capture and eventually reproduce. A microphone is in an extremely complex sound field, of sounds from instruments themselves (think of a piano) and with the addition of reflections (not less than a floor). The effect is everywhere in the physical/acoustical world. The way to deal with it in loudspeakers is in the selection of drivers and system design, and obviously some do it better than others. Technical perfection may be elusive, but viewed from the perceptual perspective, evidence is that perfection is not a requirement. Until one gets seriously involved with double-blind evaluations - i.e. almost nobody in the history of audio - it is possible to express all manner of opinions and to lose sleep over real physical phenomena that two ears and a brain take for granted.

 

[Nowhk]

Only if you deliberately seek out weird, anachronistic speakers (most of them!). If you only listen to speakers designed to be objectively neutral, we find...

We also find this... and differences in rooms, and position, and mood, and so on

Basically, we experience sounds differently every time, as everything in life.
There are too much variables that play a role into it.

The input signal in your heads will always differs, as well the mechanism that will process it.
We are complex

------------------------

Take you setup and play a records with your settings, right now. Now, make it possibile to switch gears, rooms and position in the room immediatly (feel free to choose a pro treated room and even more pro audio speakers as you need ). Than:

A - Would you claim you won't notice differences in the perceived sound? I don't think so.
(or)
B - Would you claim your brain, giving some times, will adapt to It? Than the industry of gears is too much overrated, since (if thats true), a basic pro setup and fixed room advanced requirements will be enough for everyone in the planet, ever (damn, we adapt to it...)!!

 

[j_j]

Well, if you measure speaker response on a cochlear-filter length measure, you don't need the downward slope, I think.

But it's hard to do.

 

[Floyd Toole]

It is always intriguing to involve physiological/psychoacoustic factors into measurements. It has been useful in the past. If I understand your notion correctly, you would apply such a metric to in-room measurements. While it is true that humans respond to much - not all - of the complex sound field in rooms, the awkward reality of such measurements is that two ears and a brain respond very differently than an omnidirectional microphone. So, playing with time windowing (applying technical or human-based rules) or spectral weighting (applying technical or human-based rules) yields results that do not include evidence of the directions from which reflections arrive, nor, within the time window, evidence of time of arrival. To human listeners both factors are very real information. Technical measurements may show evidence of "comb filtering" from a lateral reflection, which implies an effect on timbre. However, the reality is that the human hears normally-innocuous, if not beneficial, spaciousness.

Returning to first principles, it is clear that a double-blind listening test incorporates all physiological and psychoacoustic factors related to hearing, while suppressing or eliminating non-acoustical factors. Answers from these tests have been used to guide examinations of measurements over my 50 years of research and the key findings showed up in the very first tests I did back in 1966 (shown at the beginning of Chapter 18 in the third edition of my book). Since then the evidence has simply become more solid, supporting a remarkably simple fact - human listeners place the direct sound (the anechoic on-axis and listening window) frequency response of the loudspeaker in a dominant position with respect to timbre perception, and obviously with respect to spatial perceptions because it is the first in the series of temporal events. These measurements must be not less than about 1/20-octave resolution to show audible resonances reliably - critical bands or ERBs are insufficient.

The attached figure is Figure 13.1 from my book, and it shows that loudspeakers that elicited high ratings in double-blind tests in my 1986 JAES paper, while I was at the NRCC and in recent evaluations at Harman, 30 years apart, exhibit the same visual correlation - flat and smooth direct sound is preferred. Over the 30 years engineering has simply improved. The frequency-dependent directivity of today's forward firing loudspeakers is similar to those of the past, so the "early-reflections" data are similar, and these turn out to be reliable indicators of in-room steady-state measurements: i.e. downward tilting. The obvious conclusion is that smooth, flattish direct sound and smoothly changing indirect sound can only exist in loudspeakers without audible resonances - i.e. timbrally neutral sources. Transducers are minimum-phase devices. In the mulit[ple-comparison method employed in those double-blind tests it would appear that listeners recognize audible colorations associated with the loudspeakers, that are not in the programs, and reject them. The least colored loudspeaker is perceived as being the best one.

Evidence so far indicates that conventional in-room measurements are not reliable indicators of perceived sound quality, and for the reasons noted above, may not have that potential. Time-windowed measurements can get one closer to anechoic direct sound data, but the limited frequency resolution at low frequency ultimately limits usefulness. The consolation is that comprehensive anechoic data, suitably processed and presented (the spinorama), offer considerable insight, as is explained in great detail in my book. Can physiological/psychoacoustic factors add value to these analyses? Obviously yes, because such data are already part of the analysis - the audibility of resonances being one of our (Toole and Olive) contributions. Applying simple interpretations of this information Olive was able to show very high correlations between real (from double-blind tests in a normal room) and predicted (from anechoic spinorama data) sound quality ratings on loudspeakers - 0.86 for loudspeakers of all sizes and prices, and 0.996 for similar bandwidth bookshelf speakers (p < 0.0001). This is not accidental.

The big missing ingredient in the analysis of loudspeaker performance is a measure of non-linear distortion that reliably correlates with perception. The common harmonic and intermodulation metrics are almost useless. That is where perceptual (simultaneous and temporal masking) factors are absolutely required in the metrics.

Thanks for entering the discussion.

 

[j_j]

The big missing ingredient in the analysis of loudspeaker performance is a measure of non-linear distortion that reliably correlates with perception. The common harmonic and intermodulation metrics are almost useless. That is where perceptual (simultaneous and temporal masking) factors are absolutely required in the metrics.

Thanks for entering the discussion.

This in particular is where having an accurate analysis in loudness across frequency, masking across frequency, and error spectra, all come in.

Although it's difficult to show in a graph, this is where several things show up:

1) resonances at higher frequencies that smear the cochlear response (time domain spreading)
2) differences in time delay across frequency (big issue for percussion and other "attack" kinds of signals)
(note these two are still "linear system" issues)
3) Different phase response with level (hardly linear, but a big problem with some kinds of drivers, and often something that is different across a set of "identical" drivers ).
4) Error spectra that result in "new frequencies" being generated that are substantially below (i.e. difference tones) the original signal spectrum, since "upward spread of masking" arises directly from cochlear dynamics, but "downward spread of masking" simply doesn't exist in a normal cochlea response. Again, one has to measure the loudness of the error spectrum, not the energy of the error spectrum. Yes, it's "kind of possible" (obviously, or many other things most people use every day would not work) but both measurement and analysis must be on a window defined by cochlear filter response time/frequency tradeoff across frequency, or one can easily both overemphasize and completely miss problems.
5) Phase coherence (the ear is more or less minimum phase) that is not minimum phase or constant delay. (And constant delay is "ok" only for some kinds of filtering or shaping)

This is all on top of a good match between direct and 360 response above a few hundred Hz. (I hesitate to pick a number there, the dependency on room as well as speaker is best described as enormous.)

 

[j_j]

Until one gets seriously involved with double-blind evaluations - i.e. almost nobody in the history of audio - it is possible to express all manner of opinions and to lose sleep over real physical phenomena that two ears and a brain take for granted.

In particular, the time sensitivity of the cochlea simply buries many reflections in the 1 to 10 millisecond range, which makes sense from an "application" viewpoint simply because that's what we (in standard acoustic environments) must cope with all the time.

More complex "errors" that are not typical of normal room/building/outdoor acoustics, on the other hand, can stand out like a searchlight in a cave.

 

[Floyd Toole]

Not mentioned in my earlier post #311 is the somewhat surprising fact that the same results have been obtained in a wide range of listening rooms. As JJ points out: "This is all on top of a good match between direct and 360 response above a few hundred Hz. (I hesitate to pick a number there, the dependency on room as well as speaker is best described as enormous.)" And it is, if one has no opportunity to adapt, to become familiar with a listening space. With even a short time to adapt, everything changes.

Section 7.6 elaborates on the significant abilities of humans to "listen through" rooms to identify subtle audible properties in voices and musical instruments in live experiences, and loudspeakers in reproduced experiences. Even speech intelligibility improves with adaptation, requiring little time. We are able to separate properties of the sound sources from properties of reflective spaces. The elaborate experiment described in Section 7.6.2 is especially persuasive. These tests included changes in bass response, which account for about 30% of one's overall sound quality rating, and which can be enormous. Part of the adaptation to each of the rooms had to include bass, making the process even more impressive.

If we could not do this perceptual separation of source and space, everyday life would be extremely complicated. It happens well above the peripheral hearing system; it is cognitive and binaural, and memory is involved. Interesting stuff.

In Chapter 17 I discuss hearing loss, and in section 17.3 address a disturbing recent discovery: hidden hearing loss. In addition to the well known elevation of hearing thresholds as a function of "wear and tear" in the cochlear mechanisms, humans with - and without! - such losses can exhibit a disability in binaural hearing. We are less able to distinguish multiple sources in space, and one has to assume, aspects of space itself. This being so, the suspicion is that our ability to adapt to listening spaces, and to separate sources from venues, deteriorates as well (anybody have problems carrying on discussions in restaurants?). Hearing loss itself is commonly associated with degraded peripheral apparatus, but this new factor seems to involve more central auditory processing. Future developments will be interesting to follow.

 

[j_j]

There are at least two parts to the failure of binaural hearing. One is an inability to deal well with ITD's (interaural time differences), some people who very different hearing ability in two ears can still do this, other people with "normal" monaural hearing can't. This is deeper into the CNS than I prefer to get.

The other is when the loudness mismatch between the two ears is larger than the ability to adapt inside the CNS. This is a peripheral (i.e. ear related) problem.

One can have any combination of either or both.

 

[Thomas Lund]

We naturally distinguish between direct sound and reflections, based on experience/training, physiology, and movement. However, some reflections actually influence a human listening experience more profoundly than we tend to realise. The topic has been covered in papers; and we found the same in hi-end immersive reverb/ER design, with handles for manipulating level, phase, azimuth, elevation and diffraction effects.

A loudspeaker's definition in space therefore influences the integrity of the direct sound and of the reflections. However, just how coloured reflections can be from a “displaced” three-way system, even when on-axis response is ok, didn’t really occur to me before doing systematic tests recently in an anechoic room, and played listening examples at AES 145.

If that was the phenomenon discussed in previous posts, maybe we should avoid the term “point source”, because omni-directional isn’t normally the goal in loudspeakers. “Coaxial radiation with controlled directivity” would be more appropriate, at least considering pro audio, and complementary to a microphone.

 

[j_j]

I'm not sure why the emphasis on "naturally", surely the cochlear dynamics came about by natural processes, in particular the utility of the animal to locate sound and understand the local soundfield, but there's nothing mystical about it.

But does the radiation pattern of a speaker matter, oh, yes, it does.

 

[andreasmaaan]

2) differences in time delay across frequency (big issue for percussion and other "attack" kinds of signals)

5) Phase coherence (the ear is more or less minimum phase) that is not minimum phase or constant delay. (And constant delay is "ok" only for some kinds of filtering or shaping)

Hi @j_j, would you mind elaborating on these two points please?

If I read you correctly (which I’m not sure I have tbh), only linear phase loudspeakers are acceptable in your view? Is that correct?

 

[j_j]

Hi @j_j, would you mind elaborating on these two points please?

If I read you correctly (which I’m not sure I have tbh), only linear phase loudspeakers are acceptable in your view? Is that correct?

There's not a short answer. To some extent, constant delay (let's call it that, linear phase is kind of misleading) is important, but it doesn't have to be exactly constant delay. Phase inversion in some frequencies matters, in others it does not as long as it doesn't change the signal envelope.

There is a lot to be said, but it requires a blackboard and time.

A look at binaural masking level depression might be a starter even though we are talking here about monaural issues.

 

[Cosmik]

I'm not sure why the emphasis on "naturally"

I would emphasise that it is natural in order to highlight the absurdity of trying to correct something that isn't necessarily broken.

There could be two speakers, one of which is constructed from several transducers placed widely apart, some of which may be delayed, or emitting anti-phase etc., and we could have a device with much simpler behaviour that resembles a directional, acoustically-small transducer. We have a situation where people could look at in-room measurements for these two devices and declare the first speaker superior to the second on the basis that it gets closer to some arbitrary target curve.

This would be the result of two philosophies:

1. The in-room curve is paramount. Speaker and room are a system whose combined frequency response should match that of the recording as close as possible - or some target modification thereof (the famous downward slope). This philosophy is a denial of the idea that the listener naturally separates direct sound from room (using potentially unknown mechanisms that lie buried in the brain, also using head movements and binaural processing) and instead asserts that the listener only hears 'smoothed composite frequency response'.
2. The speaker should be neutral. Flat frequency response, no phase/timing distortion, and uniform dispersion at all frequencies. There is no need to measure its in-room curve because we're not going to be changing it, anyway. This philosophy acknowledges the human's natural ability to separate room from direct sound. (It doesn't say that the human doesn't hear or acknowledge the room sound, but it does say that you can't improve your room by modifying the direct sound).

One philosophy emphasises 'natural' and the other denies it.

A red herring is that real speakers do not have perfect dispersion therefore the in-room curve can be used to correct the real speaker in a way that an anechoic measurement couldn't. It implies that imperfect dispersion is 'no biggy' because you can always meet a target curve by fiddling with EQ - and the listener only hears the smoothed composite frequency response.

The non-red herring answer is to say that spin-o-rama type knowledge, or simulation, or predictions based on speaker geometry can be used to help pre-compensate the speaker's response - it cannot be a correction; uniform dispersion will always be superior. Reduction in the number of variables to, perhaps, a single dimension can then allow the final trimming to be performed by ear. Again, no in-room curve would be required or, indeed, be of any interest.