How accurate is Klippel?

[Marin]

Ok guys is this the same speaker?
From the pictures it looks like they are, but the frequency response is really "different" so to speak.
Especially sound power?

In mine opinion Armin has much more resolution than Erin's corner...
Same bumps and dips, same resonance at 300 Herz from free resonance of the box?
Also this looks bad if we look flat curve's from Ascend and Ascylab
But that equality comes at a high price.
You losse sensitivity and that speaker is impossible to EQ.
No headroom what so ever.
Also I'd like to ask you folks that have good buid speaker, from the designer that has some values and reputation.(Karl Fink, Andrew Jones,...)
In my opinion only they has the knowledge of how to do it with a few dollars.
So my main question is for the happy owners of this beautiful speaker at affordable price and I can be happy with that purchase even if they have some bumps and dips..with that How audible is that "panel resonance" at 300 Hertz?
I have same thing at 190 and I can't hear a thing, even if I get like 1 foot from panels?
In interview with Andrew Jones he said it was better to leave them down as low as possible where human hearing was less sensitive.
And you have much bigger problem than that, and that is the room mods that prevail down there.

 

[Mort]

You might want to break this down into one or two questions. To me it's not clear what you are asking. Are they the same speaker? yes. If you are saying the FR looks different it's within the margin of error.

 

[Marin]

You might want to break this down into one or two questions. To me it's not clear what you are asking. Are they the same speaker? yes. If you are saying the FR looks different it's within the margin of error.,

Yes within limits of margin of error, but still better resolution with Amin's measures.
Yeah you are right about so many questions,I guess the main one is that panel resolution and how audible it is?
I am at point at butcher my speaker and apply thick layer of bitumen and some reinforcements in the box.
Whoud it help and if so only for my eyes not ears,because as I said I don't hear anything at that frequency.
Funny thing is that I have slight drop due to reflection from the floor right there but it's quite narrow....

 

[NTK]

Relative to measurements of electrical signals, the accuracy of acoustic measurements are not that high. Agreement to ±1.5 dB is considered good. And that is without considering unit to unit variation of the test pieces.

 

[Mort]

Erin is probably smoothing.

 

[NTK]

Here is a post by Erin showing the effects of the mic collision protection cage of the NFS and different methods of mounting the mic on the measurement results.

Erin's Audio Corner gets a Klippel NFS!

The biggest concern I had once I set up the NFS was the microphone boom as well as the cage. The cage is a safety feature. It isn't intended to be used on the microphone at all times (not my opinion; Christian Bellman verified this when I had as Skype session with him and asked about it...

www.audiosciencereview.com

 

[amirm]

Erin is probably smoothing.

That is what I notice all the time when I look at Erin's measurements. I know early on when he first got the Klippel I mentioned how I was measuring as far as sweep resolution. I recall him pushing back saying that made testing longer which it naturally does. But I explained that it was worth the extra time. I don't know where he ended on this. And whether he has higher smoothing than 1/20 octave I use for final display.

 

[Marin]

What?
Only 1/20 octave smoothing.
More like 1/48 or more looking with REW software.
So you need more time to get that resolution.
I think it's really worth a little more time to see some resonances.And I and hole community appreciate work you doing.
I don't want to insult Erin's work and how much it has contributed to our hobby in terms of transparency.
But lately he hasn't been measuring the near field, which in my opinion is mandatory, and only normalized reflection graphs, plots...
Look here Klipsch Heresy that he measured and posted here a few years ago,and on his website now.Three times fewer measurements

 

[voodooless]

I'd be more worried about the diffrence after 7 kHz. Erin's measurement has a major bump of almost 4 dB past 10 kHz. My guess is that this is more due to sample-to-sample variation than the measurement system.

 

[dominikz]

Ok guys is this the same speaker?
From the pictures it looks like they are, but the frequency response is really "different" so to speak.
Especially sound power?
View attachment 511483In mine opinion Armin has much more resolution than Erin's corner...
Same bumps and dips, same resonance at 300 Herz from free resonance of the box?
Also this looks bad if we look flat curve's from Ascend and Ascylab
But that equality comes at a high price.
You losse sensitivity and that speaker is impossible to EQ.
No headroom what so ever.
Also I'd like to ask you folks that have good buid speaker, from the designer that has some values and reputation.(Karl Fink, Andrew Jones,...)
In my opinion only they has the knowledge of how to do it with a few dollars.
So my main question is for the happy owners of this beautiful speaker at affordable price and I can be happy with that purchase even if they have some bumps and dips..with that How audible is that "panel resonance" at 300 Hertz?
I have same thing at 190 and I can't hear a thing, even if I get like 1 foot from panels?
In interview with Andrew Jones he said it was better to leave them down as low as possible where human hearing was less sensitive.
And you have much bigger problem than that, and that is the room mods that prevail down there.

For a comparison of two random samples of a budget loudspeaker model (where sample-to-sample variation is likely significant) I'd say the agreement is pretty amazing:

(link to source)

If we look at a more premium speaker, here's a comparison of Erin's vs Amir's measurement of Neumann KH80:

(link to source)

Can't really get much closer than that! And these are again two different random loudspeaker samples.

Note that other measurement method results align well with Klippel NFS results too, here's an example of a a quasi-anechoic measurement vs NFS (so two different speaker samples measured with two completely different measurement methods):

(link to source)

It is important to remember that sample-to-sample variation is a real thing, and that there is also always some possibility of human error (though in my understanding the NFS reduces the chance of accidental measurement error significantly).
But in general, in my experience loudspeaker measurements are extremely repeatable and reliable and the NFS appears to be the most accurate method for loudspeaker measurements that we have available.

 

[amirm]

What?
Only 1/20 octave smoothing.

CEA/CTA-2034 standard mandates this resolution:

"4.3 Measurement Resolution and Calibration
The measurement resolution shall be no wider than 1/20 octave. The specific bandwidth, resolution and test stimulus used should be documented."

The research that some of the specification was based on used 1/20th octave and hence the reason it is specified such.

More like 1/48 or more looking with REW software.

Careful. Your sweep length sets the ultimate resolution. That is just the display resolution (as is what I am writing about).

That aside, more is not necessarily better. Depending on what you are analyzing, resolution can be down to 1/6th Octave.

 

[Marin]

That aside, more is not necessarily better.

Yes thanks for explaining the standard.
So 1/24 is resolution of your Klippel machine is set to.
What whoud you say is resolution of are ears.
Interesting Full HD(1080p) standard is set based that humans can't see difference from 1 metar away on 42 inch screen.

 

[Marin]

For a comparison of two random samples of a budget loudspeaker model (where sample-to-sample variation is likely significant)

You would be surprised how close sample buy sample can be thanks to today tehnolgy conveyor belts and templates for drivers,moulds itc.., but the probability from sample to sample is higher for more expensive speakers made in smaller batches.
But again it's all about how many are strict limits and how many can just be thrown away back in production because they haven't passed rigorous measurements... again more is better....

 

[Marin]

For a comparison of two random samples of a budget loudspeaker model (where sample-to-sample variation is likely significant) I'd say the agreement is pretty amazing:
View attachment 511627
(link to source)

If we look at a more premium speaker, here's a comparison of Erin's vs Amir's measurement of Neumann KH80:
View attachment 511628
(link to source)

Can't really get much closer than that! And these are again two different random loudspeaker samples.

Note that other measurement method results align well with Klippel NFS results too, here's an example of a a quasi-anechoic measurement vs NFS (so two different speaker samples measured with two completely different measurement methods):
View attachment 511629
(link to source)

It is important to remember that sample-to-sample variation is a real thing, and that there is also always some possibility of human error (though in my understanding the NFS reduces the chance of accidental measurement error significantly).
But in general, in my experience loudspeaker measurements are extremely repeatable and reliable and the NFS appears to be the most accurate method for loudspeaker measurements that we have available.

Thanks for pictures and I agree with you 99%

 

[Marin]

For a comparison of two random samples of a budget loudspeaker model (where sample-to-sample variation is likely significant) I'd say the agreement is pretty amazing:
View attachment 511627
(link to source)

If we look at a more premium speaker, here's a comparison of Erin's vs Amir's measurement of Neumann KH80:
View attachment 511628
(link to source)

Can't really get much closer than that! And these are again two different random loudspeaker samples.

Note that other measurement method results align well with Klippel NFS results too, here's an example of a a quasi-anechoic measurement vs NFS (so two different speaker samples measured with two completely different measurement methods):
View attachment 511629
(link to source)

It is important to remember that sample-to-sample variation is a real thing, and that there is also always some possibility of human error (though in my understanding the NFS reduces the chance of accidental measurement error significantly).
But in general, in my experience loudspeaker measurements are extremely repeatable and reliable and the NFS appears to be the most accurate method for loudspeaker measurements that we have available.

Wow this looks like like a completely different Twitter,or tuned differently

Now I am glad that my speaker is not messured by Klippel,he he...))) and probably never will how old and cheap it is...
Strange how much we rely on are sight and how eye-catching this is, but probably just little bit more pronanced tss,tss from drum cymbals and air considering how little music and its SPL are at that frequency.

 

[pablolie]

As a short answer to the original question, Kippel states the accuracy ranges from +/-0.1dB (under "ideal circumstances") up to +/-1dB. The latter is right at the border of audibility under very ideal circumstances with steady sine waves, particularly in the presence region.

I would never sweat over 1dB differences in any gear, the most ideal home room probably easily produces +/-2-3dB variabilities.

 

[dominikz]

You would be surprised how close sample buy sample can be thanks to today tehnolgy conveyor belts and templates for drivers,moulds itc.., but the probability from sample to sample is higher for more expensive speakers made in smaller batches.
But again it's all about how many are strict limits and how many can just be thrown away back in production because they haven't passed rigorous measurements... again more is better....

In my experience measuring speakers large-scale production is not a guarantee of good sample-to-sample consistency. Have a look at this link for an example.

Wow this looks like like a completely different Twitter,or tuned differentlyView attachment 511676Now I am glad that my speaker is not messured by Klippel,he he...))) and probably never will how old and cheap it is...
Strange how much we rely on are sight and how eye-catching this is, but probably just little bit more pronanced tss,tss from drum cymbals and air considering how little music and its SPL are at that frequency.

Do note that this is above 10kHz, where the wavelength is so short (between 1.7cm and 3.4cm) that several factors can influence the measured response.
Tweeter sample variability will IME be the biggest in this part of the spectrum, measurement microphone responses are usually variable here as well (should be taken care of by calibration, however), and it is where reflections from microphone casing/stand/clip can be visible.
Personally I wouldn't be surprised if measurement distance, temperature and humidity play a role as well.

 

[MAB]

Yes thanks for explaining the standard.
So 1/24 is resolution of your Klippel machine is set to.
What whoud you say is resolution of are ears.
Interesting Full HD(1080p) standard is set based that humans can't see difference from 1 metar away on 42 inch screen.

The comment is about 1/24 octave smoothing.
Which is different than 1/24 octave resolution.
In fact, the near-field scanner's resolution is higher than the smoothing used.

 

[Marin]

The comment is about 1/24 octave smoothing.
Which is different than 1/24 octave resolution.
In fact, the near-field scanner's resolution is higher than the smoothing

It may seem like I'm mixing the two, but there's clearly a difference.
Can you explain a layman like me)

 

[MAB]

It may seem like I'm mixing the two, but there's clearly a difference.
Can you explain a layman like me)

The microphone used by a near-field scanner like a Klippel has much higher frequency resolution than 1/24th of an octave. By much higher I mean many orders of magnitude higher.

The measurement microphone also has much higher amplitude resolution than 1 dB, again by orders of magnitude.

For example, the room in which the measurement is made has sources of uncorrelated noise much larger than the mic's resolution. So the signal is smoothed by integrating the signal over a frequency range of 1/24th of an octave. This is a common practice with measurements, for example simple voltmeters have smoothing so that the needle isn't jumping all over the place while trying to measure voltage. The smoothing is a consequence of high resolution, not the actual resolution of the measurement device.