Have you tried such measurements, taken at the listening position where an auto-eq system would expect you to place the microphone ? What reflection free period can you obtain ? 1ms ? 1.5ms at the very best if you removed the sofa and coffee table ? (Who using auto-eq would do that..) Not enough for accuracy below mid treble regions, definitely not for midrange.
They can't do it right, (other than bass frequencies) because it's not possible to do it right, for all the reasons I previously stated.
Insufficient reflection free time and an inability to draw conclusions from a single point measurement being the two main issues. Neither of these issues is going away in a listening room with in-situ speakers. To improve high frequency response flatness of a speaker requires properly conducted anechoic or pseudo-anechoic measurements (with sufficiently long window times) on and off axis. Not placing a microphone stand on the sofa and pressing "go" on an auto-eq system.
Last edited:
I'm with Simon here. In order to do meaningful corrections in the mids and treble you have to know what the speaker is doing. Getting accurate measurements in a living room is hard enough for someone who has experience measuring and designing speakers - for automated room-correction systems used by a layman it's pretty much impossible.
EDIT: Simon has expanded his post a bit since I pressed the reply button. He basically says the same.
EDIT: Simon has expanded his post a bit since I pressed the reply button. He basically says the same.
Last edited:
1ms will result in those gated measurements being valid from 1kHz upwards. Audyssey or Dirac measure multiple points. There are several papers that describe the process. Send me a PM if you're interested.
In the end all that matters is how we perceive a sound field. Do we only hear the direct sound? Probably not.
Amen to that. Believe me, I would love to be able to get accurate, high resolution measurements from midrange upwards with my speakers left in-situ.
It annoys me no end that I have to push the furniture to the edge of the room and lug the speaker onto to a stand near the middle of the room to maximize reflection free window time to get a meaningful measurement down to about 500Hz or so.
If I could do it with the speaker in place, I would, but it's not possible even with the microphone at 1 metre let alone at the far end of the room. An auto-eq system is faced with an impossible challenge at middle and high frequencies.
Well, "Convictions are more dangerous enemies of truth than lies".
Read the published papers, listen for yourself, do measurements.
Yes, 1 ms is sufficient to get meaningful results, but frequency resolution is terribly poor.
You are right. I'd like to read those papers, I'll send you a PM!
I don't think the studies support this. I haven't seen anything that says "aim for this axial response and this power response. If you can't achieve one then achieving the other is still good." The studies instead show that axial response (or in-room early response) is a very important factor and power response is a very poor factor. The radiated power can have holes and be generally non flat, and the speaker will still be highly ranked.
This is not to say that radiated power response is irrelevant and that there aren't boundaries beyond which it will degrade the perceived response. If it is too high in level then the system sounds too bright. If it has peaks they can be audible. So we know that power response has some influence and the possibility is there that two speakers have the same axial response and still sound different.
What I object to is anybody designing a system based on power response as a priority over axial response. I think that is clearly unjustified. This also means that constant directivity, in and of itself, is not a worthwhile goal. (But, smooth response on and near off axis, which may come with constant directivity is a worthwhile goal.)
I wasn't saying your JBL examples would sound bad. I assume they would sound good. I'm just saying that Toole's test showed that they could have significant power response dips at the crossover points and still sound good. His data is very clear on that. Sean's later tests with factoral analysis show the same: that axial response flatness and smoothness are highest ranking factors.
I have played with the Audesey units. I evaluated their approach for NAD. It was a room curve (steady state) based approach that has all the shortcomings of any room curve correcton. It can't seperate the early sound from the late, so a non-flat room curve must be used as a target. This is a hit and miss approach.
David S.
Markus,
This is a common fallacy in windowed FFT measurements. "Valid" ? Maybe, technically. Accurate ? No way.
People seem to forget (or be unaware) that the window time doesn't just affect the lowest "valid" frequency, as if there were a hard cut-off where everything above it is accurate, it also affects the measurements frequency resolution throughout the entire frequency range.
What is the frequency resolution with a 1ms gate time ? Yes, that's right, 1Khz. This means the entire measurement is smoothed with a 1Khz wide window, and any variations narrower than 1Khz (such as resonances) will be smoothed out. From 1-2Khz you effectively have octave smoothing that will remove all detail such as resonances, with the effective smoothing gradually becoming narrower per octave as frequency goes up.
If you don't believe this is a serious problem, look at the example I've attached below.
Red is the response of a driver with a 3.635 ms window time, while yellow is the exact same measurement with a window time of 1.333ms - the smallest that ARTA will allow. Both are within the reflection free period for the measurement set up.
The time-bandwidth limit for 1.333ms is 756Hz, which is even more generous than your 1ms/1Khz example. (278Hz for 3.635ms)
By your reasoning everything above 756Hz is valid. It's clear from looking at the results that there is a severe loss of resolution for at least 2 octaves above the time bandwidth limit, and the driver resonance between ~1.4-1.8Khz has been completely smoothed away by the 1.333ms window period. (In fact even the longer window period is still too short to accurately identify the centre frequency and width of the resonances)
Without a "longer than necessary" window time you would not even be aware a resonance was there, so how could you correct for it if you were trying to apply EQ.
Notice that even at 4.8Khz a narrow deep notch is significantly smoothed by the shorter 1.333ms window period.
As a rule of thumb I don't consider anything within 2 octaves of the time bandwidth limit to be "accurate". A 1ms window time has no hope of providing sufficient frequency resolution to correct for narrow band deviations in response at anything below approximately 4Khz, so I stand by my original assertion.
Windowed measurement techniques are an extremely useful thing but people need to be aware of their limitations, and their poor frequency resolution with short window periods is an under-appreciated source of significant error. What you can't measure, you can't correct.
Attachments
Last edited:
Oh dear, bitten by the dreaded auto-refresh on the forum's maintenance page
I've learnt to copy my messages to the clipboard around this time of day as I've lost posts to the forums maintenance mode too many times
Actually, ARTA limits the number of samples to a minimum of 128. You have been using a 96k sampling rate. With a 192k you can do shorter gates, but not all sound cards allow it.
Yes you're right the limit is 128 samples, and I do use 96Khz for all measurements.
A simple way to trick it though is simply move the start pointer further back
That's probably not true for every room, that's why I asked if the room in the test was treated. A treated room will inevitable change the effects of power response.
What makes you so sure it's only steady state? Just because they use target curves? There're also differences how the various incarnations of MultEQ work. While in XT the filters have a lot of hair at higher frequencies, XT32 generates very smooth curves that obviously don't try to fix room problems. By the way, they use non flat room curves to better match the acoustics of theaters to a home environment. There's also a flat curve that can be selected.
I've learnt to copy my messages to the clipboard around this time of day as I've lost posts to the forums maintenance mode too many times
Me too
I'm fully aware that frequency resolution is affected. In our 1ms example it becomes 1000Hz which is pretty coarse. But, then again, we're dealing with (well) engineered and finished products and all we want is to optimize the way sound is perceived (think Bark scale).
I see no need to get embroiled in the "Toole results aren't valid because I don't like his room" debate.
Yes, non-flat target curves are a defining feature of steady state approaches. I also used the unit and it was obvious that it had a steady state approach. I also had a discussion with Chris the designer. Okay?
This isn't about cinema re-EQ but the well known issue that steady state corrections can't use a flat target without sounding too bright (sounds like another reason to be weary of emphasizing power response).
David
I don't see the point in talking about power response without looking at what the room does to the sound.
And he did give you a detailed answer on how they assess the impulse response? He refused to give me that information.
Its both - the loudspeakers power response is the source and the rooms power response is the sink. The steady state SPL in the room is what lies in the middle. They both need to be correct.