Skip to main content
Topic: Loudspeaker measurements and ABX (Read 5368 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Loudspeaker measurements and ABX

HA being a resort of objective listeners, I would appreciate to have comments about following test.
In a standard living room with no special acoustic treatments :
- one louspeaker A is reference (in the sense of an anchor)
- another different loudspeaker model B is placed nearby

phase1 : measure and EQ B so to have the same curve(s) as A
phase2 : listening comparison of both loudspeakers

Variation on the method
phase1 : measure and EQ both loudspakers to have the same curve(s)*
phase2 : listening comparison of both loudspeakers

Ideal measurement + EQ method is when : measurements are the same <=> sound is the same
"Same curves" is the tricky part : what measurement+EQ method gives the best results, the most similar sound ? anechoic on-axis, in-room nearfield, RTA at main listening position, multipositions averaged, psychoacoustically gated,...?
Here is the setup :

 
Some notes :
- this is only a comparison on linear distortions (frequency response and directivity) but those parameters are known to be the most important
- this method is not an absolute listening test, it is a comparison so ABC/HR or ABX tests are relevant
- both loudspeaker should preferably be quite different before EQ
- loudspeaker B must have a wider frequency range than A, because positive EQ at both ends of spectrum may give distortion
- to be more discriminant, listening tests are done in mono on one loudspeakers.

One difficulty :
- both loudspeakers should be separated enough so that room influence is a bit different on each
- both loudspeakers positions should not permit position discrimination in listening test

Most will say that ABXing loudspeakers has no sense because the differences are generally too obvious.
But in anechoic conditions, there are reports of transparency, see Salmi, AES preprint1871, 1982, LISTENING ROOM INFLUENCE ON LOUDSPEAKER SOUND QUALITY AND WAYS OF MINIMIZING IT, where loudspeakers with flat anechoic on-axis response are transparent (and so sound the same !).
Here we add room influence, directivity and walls do complicate things.
Maybe with some work, we can really have different loudspeakers sound the same in a doubleblind test.
My questions are : do you think the test is valid, do you have any recommendation for improvements ?

Loudspeaker measurements and ABX

Reply #1
You can for a single point, at least assuming both have low enough distortion.  However the angular distribution of each source won't be the same, so the correction will vary with point in the room.

Loudspeaker measurements and ABX

Reply #2
You can for a single point, at least assuming both have low enough distortion.  However the angular distribution of each source won't be the same, so the correction will vary with point in the room.
If the correction is done for a single point, with no special gating, yes this correction will be optimised only at one point and may degrade at any other position. But you can do other types of correction, ie based on spatial averaging or with frequency dependant gating,... and the optimisation can then be much less sensitive to exact position.

Loudspeaker measurements and ABX

Reply #3
- one louspeaker A is reference (in the sense of an anchor)
- another different loudspeaker model B is placed nearby

What do you mean by "nearby"?

phase1 : measure and EQ B so to have the same curve(s) as A

If you're referring to on/off axes, that's not possible.

cheers,

AJ

Loudspeaker manufacturer

Loudspeaker measurements and ABX

Reply #4
What do you mean by "nearby"?
I mean near enough so that position is not discriminating in comparison tests. In my tests, I simply put one loudspeaker above the other.

If you're referring to on/off axes, that's not possible.
Yes, we all know that if we mimic one response at one angle, it won't coincide at other angles. That was not what I meant : by "curves" I presuppose nothing, it can be RTA, on-axis anechoic response, Harman's PIR, ....the goal would be to find the measurement that is the nearest to our perception.

Loudspeaker measurements and ABX

Reply #5
What do you mean by "nearby"?
I mean near enough so that position is not discriminating in comparison tests. In my tests, I simply put one loudspeaker above the other.


That can easily put the  drivers in each speaker a foot or several feet apart, which IME is easy enough to ABX with a positive outcome.

Loudspeaker measurements and ABX

Reply #6
That can easily put the  drivers in each speaker a foot or several feet apart, which IME is easy enough to ABX with a positive outcome.
Yes, minimal audible angle is about 3° at high frequencies. This means that at 4m distance, both tweeters should be separated by less than 20cm. If the filter have symetric vertical radiation (such as Linkwitz-Riley crossovers), vertically inverting one loudspeaker may help to reduce distance between tweeters.
But if the audible difference between the two loudspeakers is only due to the angle, I would be quite satisfied

Loudspeaker measurements and ABX

Reply #7
Did you ever conduct this test?

I've tried this myself in the past. I was using widebanders of the same size (3") and lowpassed them in the same range so directivity wouldn't be an issue. I built enclosures where I placed both drivers involved vertically. Internally the air spaces were separated and a highpass was chosen about one octave above Fsc so the natural rolloff wouldn't affect the response very much. I used an ordinary MiniDSP 2x4 with the 2way Adv parametric EQ filters to bring the FR as close as I could to each other in the listening space. I fed a multichannel receiver the same signal and only switched outputs via the remote to produce an A/B testing condition.

My goal was to find out how audible non-linear distortions were at different output levels but the test was unfortunately a failure (depending on how you view it) cause they sounded quite different anyway, obvious deviations which I didn't relate to nonlinearities at the time. In a more controlled environment this could perhaps be studied and researched further.

Loudspeaker measurements and ABX

Reply #8
Did you ever conduct this test?

Hello, yes I did the test but with no success yet : no success means that I can still hear differences between both loudspeakers.
I used two kind of FIR frequency response matching as explained in my first post : match one speaker to the other or match both to the same target.
In instantaneous switching, with some music, it is difficult to hear any difference. But with pink noise or even some voices, the difference is noticeable. I'll have to work more on the matching method but I'm not sure how near I can get.

Loudspeaker measurements and ABX

Reply #9
Did you ever conduct this test?

Hello, yes I did the test but with no success yet : no success means that I can still hear differences between both loudspeakers.
I used two kind of FIR frequency response matching as explained in my first post : match one speaker to the other or match both to the same target.
In instantaneous switching, with some music, it is difficult to hear any difference. But with pink noise or even some voices, the difference is noticeable. I'll have to work more on the matching method but I'm not sure how near I can get.


I did the first loudspeaker ABX back in the late 1970s, and after I considered the problems I encountered, I began to suspect that this would remain a unreachable quest. I think that Harman's speaker shuffler goes a long way to addressing the inherent problems, but I doubt that it really offers instantaneous switching:

You can read more about it in this paper:

Haqrman's speaker evaluation lab

The speaker shuffler is shown in Figure 4.

Loudspeaker measurements and ABX

Reply #10
If you had:

1) An anechoic chamber
2) Wanted results in an anechoic chamber
3) Were willing to do enormous processing on at least one of the speakers to match not only the direct power response, but also the phase response, of the other speaker

The answer is "perhaps". Distortion characteristics, etc, are going to make things difficult in that case.

Now, as soon as you introduce a room that isn't anechoic, the variation in the radiation pattern of the two speakers, which you can NOT adjust digitally with conventional speakers, is going to confuse the daylights out of everything, and the answer is "almost certainly not".

Hope this helped.
-----
J. D. (jj) Johnston

Loudspeaker measurements and ABX

Reply #11
That can easily put the  drivers in each speaker a foot or several feet apart, which IME is easy enough to ABX with a positive outcome.
Yes, minimal audible angle is about 3° at high frequencies. This means that at 4m distance, both tweeters should be separated by less than 20cm. If the filter have symetric vertical radiation (such as Linkwitz-Riley crossovers), vertically inverting one loudspeaker may help to reduce distance between tweeters.
But if the audible difference between the two loudspeakers is only due to the angle, I would be quite satisfied


The problem is worse than that, in that different positions will cause different reflections in any room with any reflections at all, AND you may also find HRTF interactions as well, quite aside from the usual measurements with 1 speaker at different positions in an anechoic situation.
-----
J. D. (jj) Johnston

Loudspeaker measurements and ABX

Reply #12
Now, as soon as you introduce a room that isn't anechoic, the variation in the radiation pattern of the two speakers, which you can NOT adjust digitally with conventional speakers, is going to confuse the daylights out of everything, and the answer is "almost certainly not"

The problem is worse than that, in that different positions will cause different reflections in any room with any reflections at all, AND you may also find HRTF interactions as well, quite aside from the usual measurements with 1 speaker at different positions in an anechoic situation.

YES!!

Don't want to rekindle any flame wars, but much of what you said are issues I raised with SO many moons ago regarding Harmans methods. My biggest peeve being the tests being in mono, in front of clustered listeners.

cheers,

AJ
Loudspeaker manufacturer

 
SimplePortal 1.0.0 RC1 © 2008-2019