PSYCHOACOUSTICS AND I

[Cosmik]

Psychoacoustics is the subject where the engineer tries and get the engineering just right for the listener. Example: A flat frequency response curve is a great starting point but tests showed that (an average of) people preferred systematic deviations from the «neutral» curve.

Question: placing a microphone at the listener's position, can I in principle duplicate a particular curve in a variety of ways? Particularly if a modicum of smoothing of the curve is allowed.
e.g.
1. EQ the sound from the speaker.
2. Move the speakers around, toe in and out, etc.
3. Modify the furnishings in the room e.g. draw the curtains and fit rugs, etc.
4. Vary the size and shape of the speaker's baffle.
5. Move furniture around, fit partial room dividers and so on.
6. Fiddle with the speaker's crossover: create suckouts by playing with the phase, change the individual driver levels, move drivers relative to each other.

If it is possible to state confidently that all setups giving an equivalent curve at the listener's position will sound the same to the listener, then the idea of preferred curves at the listener's position may be sound. But if not...

My view (based on my own experience of perambulating while listening, and a degree of logic e.g. above) is that an in-room frequency response curve is a 'dumb' measurement that does not duplicate the operation of human hearing. We hear the direct sound first then the reflections and reverberation, but a single frequency response measurement does not distinguish between them; our hearing does. And this means that 'a curve' (whether gated, or open or weighted in some way), does not correlate in any reliable fashion with what we hear.

This is not very useful information! It doesn't tell us what the ideal speaker is. There is a pretty good chance, however, that it may turn out to be beautifully simple: a flat frequency response and uniform dispersion at all frequencies; the room then just adds agreeable 'ambience' that looks terrible in a dumb frequency response curve, but doesn't affect our perception of the direct sound. People like Kii are simply building that idea as best they can, and it seems to work.

 

[mitchco]

@Cosmik Very interesting. So near field measurements then? In other words, how do you verify time alignment for example?

Wrt to why we hear what we hear in small room acoustics, what are your (and others if they want to chime in) thoughts on JJ's presentation: http://www.aes-media.org/sections/pnw/pnwrecaps/2008/jj_jan08/

 

[Cosmik]

@Cosmik Very interesting. So near field measurements then? In other words, how do you verify time alignment for example?

Wrt to why we what we hear in small room acoustics, what are your (and others if they want to chime in) thought on JJ's presentation: http://www.aes-media.org/sections/pnw/pnwrecaps/2008/jj_jan08/

Will check out the presentation. But as regards near field measurements, I think Keele has dealt with that. It works in theory for the bass, but not at higher frequencies.
http://www.xlrtechs.com/dbkeele.com/PDF/Keele (1974-04 AES Published) - Nearfield Paper.pdf

In the far field, you need a sufficiently clear space and to gate your measurements. A single (pseudo-)anechoic measurement doesn't tell you everything about the speaker, but the closer to ideal it is in terms of dispersion, the better it should be (based on a certain logic). A three-way speaker based on DSP should be able to get closer than a two-way simply because of the lack of beaming and the better matches between drivers. Baffle size(s) is also a factor. Or the ultimate control with extra drivers as we are beginning to see now. (IMO)

 

[Fitzcaraldo215]

I think we have to be careful about what these tests mean. If the speaker used (a B&W?) was passive and with its own foibles, then the 'room correction' may also be performing speaker correction. In the audiophile world, the assumption is that a "high quality" speaker such as a B&W is inherently good, but it would be interesting* to do a similar comparison using a modern DSP-based active speaker as the 'mule'.

*but not necessarily scientific

Yes, although called "room correction", the widely implemented concept really involves correction of the final resulting output of the total audio system as measured in the vicinity of and centered on the main listening position. Since electronics, wires, etc. normally contribute little to the resulting sound, it is therefore primarily correction of the speaker + room response combined. I think that is all fairly obvious.

I am not sure as a listener whether I care much about the distinction between correction of measured issues with just the speaker or of issues induced by the room, since they become inseparable in what is heard. I am happy correcting the final combined result, to the extent that such EQ tools can achieve it. Properly applied, I think they work effectively.

I think a better result can be achieved with a more nearly perfect speaker as measured anechoically. But, if I happen not to have one and being a pragmatist, I will settle for any perceived improvement I can get regardless of whether the cause was the speaker, the room or both. However, it is no panacea and it does not cure all ills.

DSP "room EQ" seems to have become popular in the era of Mch home theater. Mch music listening is my preference in a HT-like 5.1/7.1 system setup. The standard for Mch music recording and mastering is the ITU speaker layout using 5 identical equidistant speakers in a specific, defined angular layout. I have seen this in visits to a few classical recording control rooms.

However, such a setup is difficult or impossible in most homes. So, DSP can be used for simple time/distance correction by channel making the result effectively and virtually equidistant. DSP bass management can redirect all deep bass from each channel to one or more subwoofers. And, "room EQ" can "voice" the main, center and surround channels similarly in the region of the MLP by having all adhere to the same target frequency curve, better approaching the ideal of identical speakers all around, even if smaller center and surrounds are used. I find that this works pretty darned well.

 

[Fitzcaraldo215]

My view (based on my own experience of perambulating while listening, and a degree of logic e.g. above) is that an in-room frequency response curve is a 'dumb' measurement that does not duplicate the operation of human hearing. We hear the direct sound first then the reflections and reverberation, but a single frequency response measurement does not distinguish between them; our hearing does. And this means that 'a curve' (whether gated, or open or weighted in some way), does not correlate in any reliable fashion with what we hear.

As Toole points out frequently, an omni mike does not hear the same way that two ears and a brain do.

The widely accepted downward sloping target curve with increasing frequency is one adjustment for this difference between measurement and perception. These curves are estimated by experimentation on human test subjects.

Also, I believe that some EQ tools effectively distinguish between direct and reflected sound by time gating and sophisticated multipoint averaging techniques, which are more complex than the simple arithmetic mean. But, the details are buried in proprietary code.

It is also true that direct and reflected sound are distinguishable to the ear only if the reflections are sufficiently delayed. Otherwise, they are combined by the ear in the Precedence Effect, and the presence of reflections is masked while still potentially influencing perceived sound. Later reflections are perceivable as echo and reverb.

 

[mitchco]

Will check out the presentation. But as regards near field measurements, I think Keele has dealt with that. It works in theory for the bass, but not at higher frequencies.
http://www.xlrtechs.com/dbkeele.com/PDF/Keele (1974-04 AES Published) - Nearfield Paper.pdf

In the far field, you need a sufficiently clear space and to gate your measurements. A single (pseudo-)anechoic measurement doesn't tell you everything about the speaker, but the closer to ideal it is in terms of dispersion, the better it should be (based on a certain logic). A three-way speaker based on DSP should be able to get closer than a two-way simply because of the lack of beaming and the better matches between drivers. Baffle size(s) is also a factor. Or the ultimate control with extra drivers as we are beginning to see now. (IMO)

Yup up to speed on Keele. What I am specifically referencing to JJ's paper are the reasons why we hear what we hear in small room acoustics. It is all about spychoacoustics - which is very pertinent to your thought about "room correction". Do you agree with what JJ is presenting? If so, then you will see far field really does not matter so much... it is the direct sound timbre that matters most.

Wrt to measurements, what I am asking you specifically, is that do you have any measurements of any kind to share that can show your system is indeed time aligned.

As a side note, I think my signature gives it away, but having played with DSP software over the past 7 years in many rooms with many systems, along with others from around the world sharing their in-room measures, there is indeed a direct correlation to what we measure in room at the listening position and what sounds neutral to our ears. Amir has already referenced the work by Olive and Toole. Are you saying you disagree with that?

 

[svart-hvitt]

Question: placing a microphone at the listener's position, can I in principle duplicate a particular curve in a variety of ways? Particularly if a modicum of smoothing of the curve is allowed.
e.g.
1. EQ the sound from the speaker.
2. Move the speakers around, toe in and out, etc.
3. Modify the furnishings in the room e.g. draw the curtains and fit rugs, etc.
4. Vary the size and shape of the speaker's baffle.
5. Move furniture around, fit partial room dividers and so on.
6. Fiddle with the speaker's crossover: create suckouts by playing with the phase, change the individual driver levels, move drivers relative to each other.

If it is possible to state confidently that all setups giving an equivalent curve at the listener's position will sound the same to the listener, then the idea of preferred curves at the listener's position may be sound. But if not...

My view (based on my own experience of perambulating while listening, and a degree of logic e.g. above) is that an in-room frequency response curve is a 'dumb' measurement that does not duplicate the operation of human hearing. We hear the direct sound first then the reflections and reverberation, but a single frequency response measurement does not distinguish between them; our hearing does. And this means that 'a curve' (whether gated, or open or weighted in some way), does not correlate in any reliable fashion with what we hear.

This is not very useful information! It doesn't tell us what the ideal speaker is. There is a pretty good chance, however, that it may turn out to be beautifully simple: a flat frequency response and uniform dispersion at all frequencies; the room then just adds agreeable 'ambience' that looks terrible in a dumb frequency response curve, but doesn't affect our perception of the direct sound. People like Kii are simply building that idea as best they can, and it seems to work.

I think what you actually try and describe, is the perfect source. True point source is a very convincing concept, as far as I understand. True point source and omnidirectional: There you have it! Right?

 

[svart-hvitt]

FWIW: Let me tell you about my own experiences with the Genelec GLM system (DSP) and the 3-way coaxial 8351a.

It’s more a matter of apples to apples than apples to oranges. It’s not night and day when I turn DSP on-off. One friend of mine preferred without room-mic calibration. However, since Genelec only take away energy (cutting peaks), they’ve made their own system prone to loudness preference (people prefer louder)! In my ears, the bass started to bloat a bit without DSP, so I keep it on.

Given the fact that speakers and software integration are complex, I have sympathy with the Genelec approach which reminds you of Apple’s integrated approach. And then we are in the midst of the MS vs Apple discussion (with no end on sight).

This was a very short summary of my own experience with room-mic calibration. I lean in the direction of believing that DSP is valuable in lower frequencies.

 

[amirm]

My view (based on my own experience of perambulating while listening, and a degree of logic e.g. above) is that an in-room frequency response curve is a 'dumb' measurement that does not duplicate the operation of human hearing. We hear the direct sound first then the reflections and reverberation, but a single frequency response measurement does not distinguish between them; our hearing does. And this means that 'a curve' (whether gated, or open or weighted in some way), does not correlate in any reliable fashion with what we hear.

That's not the case in frequencies below transition (a few hundred hertz). The room dominates the response and what you see in measurements is what you get with with your ear to most part. Above that it gets tricky fast as you mention.

 

[Cosmik]

I think what you actually try and describe, is the perfect source. True point source is a very convincing concept, as far as I understand. True point source and omnidirectional: There you have it! Right?

I think it would turn out to be point sized, but in practice I think it would not necessarily be omnidirectional except in a small selection of rooms. I suspect that B&O are onto something with their adjustable dispersion idea, where we would find that moderate directionality is desirable, but in many rooms we would not want it too wide. Too narrow, however, and we lose desirable 'ambience' (like listening in an anechoic chamber) and reduce the size of the 'sweet spot'. Whatever the dispersion angle is, it should ideally be uniform with frequency. The idea of Kii and similar speakers is to keep it uniform all the way down into the bass.

 

[Cosmik]

Wrt to measurements, what I am asking you specifically, is that do you have any measurements of any kind to share that can show your system is indeed time aligned.

No, they are time-aligned 'by inference'. The individual drivers are rendered linear phase, and delays appropriate to their position relative to the listener are applied. Confirmation is by sweeping delays while listening to pink noise from two adjacent drivers around the listening position - the wrong delay gives comb filter-style cancellation through the crossover heard as a 'thinning out' of the sound.

Are you hinting that there is some mystery to time alignment? The idea is to make several cones move as if one. They're no doubt not perfect - a sweep-based measurement probably doesn't perfectly represent their behaviour with a genuine step signal, etc., but control using DSP is an order of magnitude better than the passive alternative..? Also, the drivers are physically separated, so time alignment only applies to one position in space. But, given that we can't have a point source, we just have to get as close as possible - IMO.

 

[RayDunzl]

a sweep-based measurement probably doesn't perfectly represent their behaviour with a genuine step signal

I found that it does... (or can)

Using a 10Hz (100ms cycle) square wave as the "step" , and comparing the in-room audio recording with the calculated from swept sine sweep, the results were essentially identical.

Let me find it... Posted here someplace...

Voila...

Top: REW calculated step response from a short (128 or 256ms) full-range sine sweep.
Bottom: Audacity in-room audio recording of 10Hz square wave

This little test greatly increased my appreciation for the sophistication of the analytical tools freely available to us proletarians.

 

[svart-hvitt]

I think it would turn out to be point sized, but in practice I think it would not necessarily be omnidirectional except in a small selection of rooms. I suspect that B&O are onto something with their adjustable dispersion idea, where we would find that moderate directionality is desirable, but in many rooms we would not want it too wide. Too narrow, however, and we lose desirable 'ambience' (like listening in an anechoic chamber) and reduce the size of the 'sweet spot'. Whatever the dispersion angle is, it should ideally be uniform with frequency. The idea of Kii and similar speakers is to keep it uniform all the way down into the bass.

I thought «real» sound was omni-directional? How do you teach instruments about directionality?

 

[RayDunzl]

I thought «real» sound was omni-directional? How do you teach instruments about directionality?

Stand in front of, and then behind, a trumpet player.

Omni directional?

To some degree.

 

[svart-hvitt]

Stand in front of, and then behind, a trumpet player.

Omni directional?

To some degree.

I think instruments need to be taught about how speakers work. According to my sources, instruments don’t know what is stereo.

Instruments misbehaving!

 

[Cosmik]

I thought «real» sound was omni-directional? How do you teach instruments about directionality?

What you have is a recording made with some static microphones that, to whatever degree the recording person captured it, contains direct and ambient sound. You are welcome to listen to the recording in an anechoic chamber on speakers (if necessary sitting very still and using BAACH type cross-channel cancellation) or with headphones whereupon it can be arranged for you to perfectly hear the recording as it was made. Sitting very close to some small speakers would also get close to this 'ideal'. This will give the perfect measurements at the listening position that everyone says they want. Directionality of the speakers wouldn't matter because you would only be hearing the direct sound.

However, in conflict with that, the way I see it is that you may wish to listen to the recording in a real room because:
(1) You want your listening to be compatible with ordinary domestic life (used to be true in, say, the 1970s, but for modern audiophiles with their dedicated listening chambers is less true now).
(2) Real added room ambience provides some dynamic ambience that overcomes the static nature of the recording; that responds to your head (micro-)movements. It also helps to bind real ambient sounds (when you talk, etc.) with the recording as opposed to the peculiar sensation of listening to a recording of a reverberant cathedral, but your own voice and movements being reverberation free.

The quantity of this added room ambience is related to the directionality of the speakers. Sure, on strictly scientific grounds, it doesn't represent a perfect facsimile of the trumpet in a room, but it does represent a static recording of something that sounds very similar to the trumpet with added dynamic ambience - ambience derived from the recording in a coherent, 'correct' physical system that our hearing is able to interpret without added effort; naturally separating the direct from the ambient, and enjoying the ambience. Speaker directionality is a way of controlling the ratio of direct to reverberant sound that is naturally correct/coherent/interpretable by human hearing. Piping in 'ambience' from some separate remote drivers wouldn't be the same because our hearing would identify that it wasn't naturally derived from the direct sound.

 

[RayDunzl]

In addition to the Calculated Step Response above...

A similar experiment with Impulse Response.

Top: REW sine sweep and calculated impulse response
Bottom: Audacity in-room audio recording
Very bottom: Signal for Audio Recording - single full scale sample

How the magical mathematics can pull a "click" out a sine-sweep "whoop" is way way beyond my pay scale...

 

[amirm]

I found that it does... (or can)

Using a 10Hz (100ms cycle) square wave as the "step" , and comparing the in-room audio recording with the calculated from swept sine sweep, the results were essentially identical.

You are speaking of different things Ray. Cosmic is talking about a sweep where one frequency is played, measured, plotted and then we move to the next measurement.

REW is different in that it plays a log sweep and captures all of its energy at once. In that regard, in addition to its ability above, it can also represent a step function which would also have the same sum total harmonics. This data is then shown in time domain which is different than a frequency sweep (even though its source was the sweep).

 

[Cosmik]

You are speaking of different things Ray.

Hi Amir. Actually I meant what Ray means - comparing a swept sine wave to a 'real' step or impulse.

 

[amirm]

Hi Amir. Actually I meant what Ray means - comparing a swept sine wave to a 'real' step or impulse.

Oh, in that case, never mind!

Then again, let me say that swept sine has far better signal to noise ratio. That is why it gets used. It is also safer on drivers and electronics.

See:
Transfer Function Measurement with Sweeps
SWEN MÜLLER, AES member,
Institut für Technische Akustik, RWTH, 52056 Aachen, Germany
AND
PAULO MASSARANI

And a really good paper by Italian professor, Angelo Farina whose method is used by REW:

Simultaneous measurement of impulse response and distortion with a swept-sine technique