Dear David, as I am the author, I will reply because there may be a misunderstanding.
Here is picture 2 of my MMM presentation, various measurements 10cm separated, we are far from +-4dB on 1 octave wide...
Picture 5 (it is not the same speakers and not the same room, measurements made over a 3x2x1m volume) shows variations very near from Schroeder's calculation which give a 95% confidence range of about +-6.1dB at 1000Hz with RT of 0.2s.
So are we speaking of same measurements ?
Last edited:
Hi Jlo,
Yes, I was going by your figure 3, that I understood to be a 1/6th Octave smoothed version of the curves above? By that you were saying that the +-4 (you referred to it as "variations of nearly 10dB") differences were from small positional variation?
If so, I believe it is still fair to ask why those particular variations should be ignored. They look like enough change to be audible. Although arrival from off to the sides might reduce their effect, is that the case here? Without knowing the particular cause it seems hard to say.
Regards,
David
Yes, I was going by your figure 3, that I understood to be a 1/6th Octave smoothed version of the curves above? By that you were saying that the +-4 (you referred to it as "variations of nearly 10dB") differences were from small positional variation?
If so, I believe it is still fair to ask why those particular variations should be ignored. They look like enough change to be audible. Although arrival from off to the sides might reduce their effect, is that the case here? Without knowing the particular cause it seems hard to say.
Regards,
David
For the variations in jlo's data to be audible, one would find that in practice moving ones head by a few inches would cause an audible change to the sound. Prof. Benade, who knew the Schroeder statistics well and always wondered why they were not audible, believed that we move our heads during listening enough to average them out. That does not seem plausible to me however.
Do we ever find this in practice? Not that anyone has ever documented it, probably because it just does not pass the credibility test. I find little audible difference in my system across the sofa let alone a few inches.
These statistical variations have been well known for decades and everyone who has looked at them has conclude the same thing - that you need to average them out, because they are not audible.
But let's not forgot, these only apply to steady state responses where there are a lot of reflections. As such, to me, they are not very relevant when one is talking about sound quality as this is heavily weighted by the early sound which does not meet the statistical assumptions. jlo's comment from Olive is interesting because it is strong support that tone color is highly correlated to the early sound. If the speaker is constant directivity then the steady state and the early sound will be quite similar.
Do we ever find this in practice? Not that anyone has ever documented it, probably because it just does not pass the credibility test. I find little audible difference in my system across the sofa let alone a few inches.
These statistical variations have been well known for decades and everyone who has looked at them has conclude the same thing - that you need to average them out, because they are not audible.
But let's not forgot, these only apply to steady state responses where there are a lot of reflections. As such, to me, they are not very relevant when one is talking about sound quality as this is heavily weighted by the early sound which does not meet the statistical assumptions. jlo's comment from Olive is interesting because it is strong support that tone color is highly correlated to the early sound. If the speaker is constant directivity then the steady state and the early sound will be quite similar.
Well it is allways fair to ask...I don't know where the variations are coming from : they are due to reflections, part from walls, ceiling and floor, part from mixing console,....
But what is obvious is that the sound quality does not change at positions separated by only 10cm. For me, those measured differences are not audible. For sure, a +4dB over an octave wide is clearly audible if it is electronically added to the signal. But when the same difference comes from a reflection, it is perceived differently at both ears and it is certainly smoothed and damped because of the unconscious small movements of the head. I remember that David Clark did a comparison, a long time ago, between reflections added acoustically and added electronically : the audibility is completely different.
I agree with you that it would be interesting to see where the measured variations come exactly from. I already did special MMM measurements such as moving only in one plane, so to avoid averaging reflections coming from one dimension. But did not get understandable results.
Another possibility would be to use directionnal microphones. Those who have a soundfield mic could record and work on the B-format.
Everything is not clear, and I would also like to better understand why the MMM method gives such good results. More work...
Maybe we don't move enough in distance but we slightly turn our head (this is know as small head-movements in sound localization). That could give us the possibility to average out the reflections effects.
Try to listen with fixed head : does it change the sound quality ?
Some years ago a family friend was over for a listen. She has incredible hearing, well past 25k. Oddly enough after a couple of songs she selected we had to stop. She made several suggestions about this and that. But when asked why we stopped the session the response was, "it hurts". At the time she was in a halo ring recovering from a horrible accident and couldnt move her head in responce to these cues.
That got me thinking and noticed that it is quite unnatural not to move the head. Kinda like the "head in a vise" but in reverse.
Its fine to say that late arriving reflections are less audible and that the "grass" we see on a curve. per Schroeder, should be averaged out. (but frequency smoothing is as good as spatial averaging, in my experience.) But when a dip is an Octave wide it is not a late reflection. That was my reference to "quefrency" earlier. Late reflections can only contribute fine comb filtering to the response. It is agreed that such late reflections will be less than a critical bandwidth and likely inaudible. If it is a full Octave wide, though, then it is not caused by a late reflection. It is in the direct sound of the speaker as an off axis issue, or it is a close time reflection such as a floor bounce. These are highly audible, as I suspect your measured source variation was too.
Even later reflections can be audible. I remember a number of times at McIntosh where you could put pink noise into a speaker and walk away from it. There was always a distinctive treble pitch component, probably from a ceiling bounce, that went up in pitch as you receded.
As to the Schroeder type "fine grass" in an unfiltered room curve, it too is audible on a steady tone. You can put 3kHz into a room and sway back and forth and hear level rise and fall, just as a microphone would measure. It won't be apparent on music because it is a steady state phenomenon and the pulsing and frequency spread of music doesn't excite it. I don't think head movement is required to make it go away.
David
You're right, the technique and the measured results are well understood.
What I don't fully understand :
with MMM, we measure and EQ mostly the direct field above a certain frequency and this is enough to get very coherent listening results.
I thought that we would have needed more of a kind of psychoacoustic windowing to include a temporal integration of reflections and also more of louspeaker power response.
Just another comment in support.
I have tried 50" gated measurement for EQ and some other methods of EQ. MMM is clearly the best for my situation. I really only have experience with one setup however so it may well not be representative of the general situation. Pink noise from all 5 speakers is audibly almost identical and there is very little difference around my listening area.
My TWs (SEAS DXT) use diffraction to distribute the sound widely. The uniformity across that width is not as uniform as a driver that does not use diffraction however. This may explain my MMM preference? I will be trying a more conventional dome for comparison at some point. My concern there is that I would then expect to have all the positional sensitivity that all others seem to experience with domes. I now enjoy a pretty solid imaging and sound balance around a large listening area.
Reading the ill effects of early diffraction has always been somewhat a puzzle to me. We add diffraction to room treatments to scatter HF to help avoid strong cancelations. When this same effect is less delayed from the direct sound (diffraction at the speaker or the cabinet edges) then we say it is bad.
The primary impact to my sound quality seems to be related to source material vs house curve. When some voices sound harsh, instruments sound dull, lacking air or lacking bass, then there is usually a correction possible with EQ/House Curve for that source. This is to point out the MMM is repeatable and if we don't lock ourselves into the irrational thought that a particular house curve must be used then I am finding it possible to get very good results using this method.
After about 30 adjustments to establish my currently preferred house curve, it works very well for my preferences, in my room, on my favored material. It is clearly still a compromise. It is far from jlo's recommended target curve for my setup and falls well below all others I have seen recommended. I can only assume the DXT's "broad scatter" may be the reason.
My point here is that there is a difference between the EQ that results from using different methods of measuring the response. We can hear differences in EQ so they will not sound alike. We then say that EQ method X works better than method Y because on that material, in that room it sounded better to the listener(s). What would have been the judgment had we used different material or a different house curve? It seems to me that the reverse conclusion could have been possible.
So far I chosen to use the method that is quick, easy and highly repeatable and with little sensitivity to setup. The house curve is then established accordingly. This doesn't solve any of the problems with EQ in general. It just makes it easier to measure and make adjustments over time.
My current 5.1 measurements using MMM:
I have tried 50" gated measurement for EQ and some other methods of EQ. MMM is clearly the best for my situation. I really only have experience with one setup however so it may well not be representative of the general situation. Pink noise from all 5 speakers is audibly almost identical and there is very little difference around my listening area.
My TWs (SEAS DXT) use diffraction to distribute the sound widely. The uniformity across that width is not as uniform as a driver that does not use diffraction however. This may explain my MMM preference? I will be trying a more conventional dome for comparison at some point. My concern there is that I would then expect to have all the positional sensitivity that all others seem to experience with domes. I now enjoy a pretty solid imaging and sound balance around a large listening area.
Reading the ill effects of early diffraction has always been somewhat a puzzle to me. We add diffraction to room treatments to scatter HF to help avoid strong cancelations. When this same effect is less delayed from the direct sound (diffraction at the speaker or the cabinet edges) then we say it is bad.
The primary impact to my sound quality seems to be related to source material vs house curve. When some voices sound harsh, instruments sound dull, lacking air or lacking bass, then there is usually a correction possible with EQ/House Curve for that source. This is to point out the MMM is repeatable and if we don't lock ourselves into the irrational thought that a particular house curve must be used then I am finding it possible to get very good results using this method.
After about 30 adjustments to establish my currently preferred house curve, it works very well for my preferences, in my room, on my favored material. It is clearly still a compromise. It is far from jlo's recommended target curve for my setup and falls well below all others I have seen recommended. I can only assume the DXT's "broad scatter" may be the reason.
My point here is that there is a difference between the EQ that results from using different methods of measuring the response. We can hear differences in EQ so they will not sound alike. We then say that EQ method X works better than method Y because on that material, in that room it sounded better to the listener(s). What would have been the judgment had we used different material or a different house curve? It seems to me that the reverse conclusion could have been possible.
So far I chosen to use the method that is quick, easy and highly repeatable and with little sensitivity to setup. The house curve is then established accordingly. This doesn't solve any of the problems with EQ in general. It just makes it easier to measure and make adjustments over time.
My current 5.1 measurements using MMM:
That's why people should probably say "diffusion" when they mean randomly scattering the sound, and "diffraction" when they are talking about coherent edge diffraction causing big dips and peaks in the directly-heard frequency response.
Here is what I think is happening. Frequency averaging is smoothing the response more and more at HFs because the bandwidth is larger (you are using 1/6 octave). At LFs the variations are smaller because the wavelengths are long (the variations cannot be shorter than the wavelength.) This is why you see a small band (approximately Gaussian I might add) in the center than has variations and nothing at either end. The direct field may dominate above a certain frequency, but this is likely to be above the band of frequencies where the uncertainty is large in your example.
I would say that there is a good reason for that. The closer the diffraction is to the original source the shorter the time lag between it and the direct sound. When it happens very early it adds to what Blauert calls "summing localization" and interferes with the image and adds coloration. Longer time lags and this effect gets summed into the reflection pattern and is not as big an issue.
I would agree that very wide directivity HF sources will sound brighter in most rooms, but you preference curve is a little more pronounced than mine.
Choosing EQ, taget curve,.... may be highly subjective (why ?) but for the measurement itself, one should only rely on a method that is metrologically founded : accurate, precise and repeatable.
This is right.
But why will the direct field will dominate above a certain frequency ? We are far above critical distance.
Last edited:
Yes, If the house curve "A" is created based on a measurement method "A" then changing the measurement method to "B" but using the same house curve "A" may well lead to a problem with sound quality. The converse is also true. The moral being to pick a convenient and consistent method.
Just another way to say it. I'll stop promoting this now.
I am currently studying the effects of early sound and late sound on our perception and getting unexpected results. I too believed that the direct field dominates perception and later sound can be ignored, leading to the typical "house curve". In standard sized rooms, though, I am finding the reverberent field spectrum to be fairly important.
As to critical distance, in many hifi situations we are sitting at or just a little beyond the critical distance. At high frequencies we are likely well within the critical distance. Don't discount the direct field, and from midrange frequencies down, the first wall or floor bounce is audible as well.
David S.
A lot of good material in the past couple of days! (I've been away)
Dave. I do agree that just because a technique glosses over something does not mean it should, or that it's not audible. The question might be - is a large number of averages over a given area a better representation of how we hear than a fixed spot?
Another basic question. Are MMM results any different than averaging multiple fixed points? One of the papers linked to attempts to answer that question.
In much larger rooms, where I typically work, the reflected sound can make a huge difference in perception. Anyone touring with a P.A. will tell you how much the venue changes the sound of the same system. It's a constant struggle. Less so with speakers of a fairly constant directivity, but still a struggle. Small rooms are different, but will also effect perception. Taking the small speakers around to different Hi-Fi shows will teach you that - fast.
I understand what you mean, but wonder how much of a problem it really is. Certainly in my present room a MMM has the same general curve as fixed, it's just smoother. With MMM I have found some problems I could hear, but had a hard time measuring, and vice-verse with fixed showing things I couldn't really hear. Still, the curve isn't wildly different and the MMM has gotten me closer to what I hear. Averaging 6 or a dozen fixed locations should do the same. I don't know, of course. how this applies to other systems and rooms. At least not yet.
Dave. I do agree that just because a technique glosses over something does not mean it should, or that it's not audible. The question might be - is a large number of averages over a given area a better representation of how we hear than a fixed spot?
Another basic question. Are MMM results any different than averaging multiple fixed points? One of the papers linked to attempts to answer that question.
With my current experimentation I can increase reverb and decrease direct sound with any step size and instant switching. While I am looking for response shifts, the much larger effect is the perception of changed space and distance. There is also a sense of resonance when the reverb level increases. Even though the response is largely the same for both cases it seems like certain frequencies ring on and give an emphasis not obvious in measured response. I would imagine this is a big part of large room PA sound.
The overriding issue is that a raw response curve is too much information. A high resolution graph shows a thousand peaks and dips, most of which we don't resolve. Yep, can't see the forest for the trees....The question is how to simplify to a curve where shape just matches perception. Spatial averaging, frequency smoothing, time windowing....those are the only choices I know of.
I don't see a large difference between spatial smoothing and frequency smoothing and certainly critical bandwidth theory suggests that we should frequency smooth to a sixth Octave or so. Toole argues against it by saying that smoothing loses sight of resonances that may be audible, but he is assuming anechoic measurements, so he has already "time windowed" with a reflection free response. My only caveat with spatial smoothing is that if we encompass areas beyond our listening position we might encounter response aberations that aren't heard or hide aberations in effect at our actual position. Clearly the ear has no sense of what the response is for places where it ain't at.
Spatial sampling (an averaging of fixed space sample points) can't be different if the number of sampling points is sufficient. This is the same as sampling of a waveform: discrete samples = continuous waveform if sufficient number of points are taken.
David
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.