Master Preference Ratings for Loudspeakers

[MZKM]

I'm confused. EQ'ing a speaker is an all-or-nothing proposition: it will affect all curves at once, because EQ applies all radiated sound equally. From that perspective, I don't understand what you're trying to say by "If you EQ the LW/ER/SP, you have no clue how it affects the curves that make them up". It's not possible to "EQ the LW/ER/SP". You can only EQ the entire speaker, and whatever EQ you apply affects all curves equally in a perfectly predictable way.

Ah, after thinking about it, you are right. For some reason I was thinking the changes would be split across the curves that make it up, which isn't true, as if you EQ the PIR curve to be +2dB @1kHZ, then all other curves must increase by +2dB, as they are all directly related, as any changes to the on-axis is applied identically to the other curves.

It would be a pain in the but though, as if you EQ the PIR, then you have to go back and apply those changes to every other curve (though easy if you have a duplicate document that applies that difference to every curve).

 

[QMuse]

as if you EQ the PIR curve to be +2dB @1kHZ, then all other curves must increase by +2dB, as they are all directly related, as any changes to the on-axis is applied identically to the other curves.

The easiest way to look at EQ is that the change is not applied to any curve in particular but it is applied to the input signal of the speaker, so ALL curves act accordingly.

 

[thewas]

Yes, it's a lot of tedious work, wish REW or would implement in the future a "batch EQ" functionality for such. Or even a Harman score calculator. Or virtuixcad.

 

[Purité Audio]

John Mulcahy posts here so you can ask him yourself.
Keith

 

[QMuse]

Yes, it's a lot of tedious work, wish REW or would implement in the future a "batch EQ" functionality for such. Or even a Harman score calculator. Or virtuixcad.

That would indeed come handy, but the main obstacle now is still a fact that @MZKM and @edechamps scoring engine accepts only Klippel's format and not REW's.

 

[pierre]

You might want to try @pierre who's already set up for computing scores from a variety of input formats.

Right now Loudspeaker Explorer strongly assumes the input is in the format Amir publishes (i.e. the Klippel export format). I'm not necessarily opposed to the idea of supporting data in other sources and formats, but that's pretty far down my TODO list. I'm not even finished implementing basic score calculation to begin with…

@napilopez

If you can parse the graphs with WebPlotDigitizer then most values are automatically computed.
There are tutorial on the internet or member XVX was kind enough to write one dedicated for Spinorama charts https://github.com/pierreaubert/spi...orial/digitalization/Digitization-Tutorial.md
if you are lazy like me, you do not need to do the 2 DI curves, the system will computed them from the others.

You have your datas in a mdat from REWS? correct?

 

[edechamps]

Yes, it's a lot of tedious work, wish REW or would implement in the future a "batch EQ" functionality for such. Or even a Harman score calculator. Or virtuixcad.

An EQ feature is on my TODO list to implement in Loudspeaker Explorer. E.g. something that would let you upload a REW EQ file and then apply it to an entire speaker, with everything recomputed from there (including score). It's going to take some time before I get to that though.

 

[thewas]

That would be a dream coming true for me, thanks again for your fantastic efforts.

 

[QMuse]

An EQ feature is on my TODO list to implement in Loudspeaker Explorer. E.g. something that would let you apply a REW EQ file and then apply it to an entire speaker, with everything recomputed from there (including score). It's going to take some time before I get to that though.

I suggest you do it with filter in standard wav 32bit LPCm mono format. Such filter can be generated with REW but also with pretty much any other EQ tool on the market.

 

[MZKM]

OK, I've had it with trying to manually calculate all the curves based on the H/V measurements and not what Klippel spits out. As you may have seen, I saw the PIR was different when manually calculated, and now I did all the curves and not just the PIR, but now the Sound Power and even the Listening window are different!

Here is the comparison using the F208:
Klippel:

Manual calculation, no ER fix:

Manual + ER fix (+/90 thru 180 for rear-wall, and the average of all the curves that make it up):

Unless Klippel is using more than the specified 70 measurements or does weighting, I don't see how the Sound Power can be as high as it reports.

Now, I could have screwed up the calculation; I can share my file if anyone wants, or if someone would start fresh on their own that may be better.

 

[edechamps]

OK, I've had it with trying to manually calculate all the curves based on the H/V measurements and not what Klippel spits out. As you may have seen, I saw the PIR was different when manually calculated, and now I did all the curves and not just the PIR, but now the Sound Power and even the Listening window are different!

Unless Klippel is using more than the specified 70 measurements or does weighting, I don't see how the Sound Power can be as high as it reports.

Now, I could have screwed up the calculation; I can share my file if anyone wants, or if someone would start fresh on their own that may be better.

I remember this post from a few months back:

FYI this is what Klippel system does:

"Power average" is also the term used in CTA-2034-A §5.2.

What formula are you using to compute the average? Note that the power average formula I just quoted is not a simple average, it's a root-mean-square (RMS or quadratic mean).

 

[MZKM]

I remember this post from a few months back:



"Power average" is also the term used in CTA-2034-A §5.2.

What formula are you using to compute the average? Note that the power average formula I just quoted is not a simple average, it's a root-mean-square (RMS or quadratic mean).

OK, that may explain it, I'll try and implement that and see what happens.

 

[edechamps]

OK, that may explain it, I'll try and implement that and see what happens.

Yeah. Though I'll admit I don't completely understand the Klippel formula myself because I always get confused with how transfer functions are squared and how that relates to power and amplitude. It might be that the frequency response data (i.e. frequency vs. SPL data) that we have is already the transfer function squared, in which case the formula amounts to a simple average in that domain (the square roots and square cancel each other out).

 

[edechamps]

@MZKM: actually, I just realized a more fundamental problem here is that the Klippel formula clearly suggests the calculation is not done in dB. You might want to try converting to linear sound pressure before averaging, then converting back. (Basically it's averaging the physical quantity, not the resulting curve.) This would definitely explain why your numbers are much lower than Klippel's: averaging in log scale (i.e. dB) puts much more weight on smaller values, compared to linear scale.

 

[MZKM]

@MZKM: actually, I just realized a more fundamental problem here is that the Klippel formula clearly suggests the calculation is not done in dB. You might want to try converting to linear sound pressure before averaging, then converting back.

Yes, I saw that too.

 

[edechamps]

I'm sure the correct interpretation of the power average formula would be obvious to people who are knee-deep into the math of transfer functions. I don't know, maybe @NTK? @ctrl? @JohnPM? Also @pierre had to look into this at some point.

 

[MZKM]

I'm sure the correct interpretation of the power average formula would be obvious to people who are knee-deep into the math of transfer functions. I don't know, maybe @NTK? @ctrl? @JohnPM? Also @pierre had to look into this at some point.

I haven't made much headway, but what I thought was interesting is this:
https://www.americanradiohistory.com/Archive-All-Audio/Archive-Audio/70s/Audio-1974-09.pdf
On page 63 it has the same weighting values as in the CTA document, except for on-axis. Also, thanks to the small Google link preview, this AES journal:
http://www.aes.org/e-lib/browse.cfm?elib=18891
Has the same value for on-axis.

It's not close either, these two state 0.002417944 and the CTA states 0.000604486.

Who is right? Who is Wrong:

Now, the CTA document states the on-axis should have little weight as the +/-10 H&V points are surrounding it, and while the two documents I found do show this, maybe it needs even less weight for the nature of the CTA 70 measurements.
So, the CTA is likely correct for this purpose, but I found it interesting.

I wish there was a weighting formula rather than just using these decimal values.

 

[MZKM]

I remember this post from a few months back:



"Power average" is also the term used in CTA-2034-A §5.2.

What formula are you using to compute the average? Note that the power average formula I just quoted is not a simple average, it's a root-mean-square (RMS or quadratic mean).

Wait, why is the weight being multiplied and then divided, canceling itself out? I just did the calculation for the on-axis and realized this (including converting to and from pascal). Unless I did it wrong, for the on-axis:
=20*log10((sqrt(((abs(0.00002*10^(Decibel/20)))^2*1*0.000604486)/0.000604486))/0.00002)

Note that the regular power average I got a lower SPL as you divide by the frequency.

 

[pierre]

Wait, why is the weight being multiplied and then divided, canceling itself out? I just did the calculation for the on-axis and realized this (including converting to and from pascal). Unless I did it wrong, for the on-axis:
=20*log10((sqrt(((abs(0.00002*10^(Decibel/20)))^2*1*0.000604486)/0.000604486))/0.00002)

Note that the regular power average I got a lower SPL as you divide by the frequency.

Here is what I use:

- all computations are on pressure (spl->pressure before computing, and pressure->spl after)
- rms Are weighted
def weighted_rms(spl):
avg = [sp_weigths_hv[c] * spl[c]**2 for c in sp_cols if column_valid(c)]
wsm = [sp_weigths_hv[c] for c in sp_cols if column_valid(c)]
return np.sqrt(np.sum(avg)/np.sum(wsm))
I didn’t check the weights themselves. I think there are defined here http://emis.impa.br/EMIS/journals/BAG/vol.51/no.1/b51h1san.pdf


we had the same discussion https://www.audiosciencereview.com/...all-speakers-reviews.11373/page-2#post-369375

i find almost the same values as Klippel.

 

[edechamps]

Note that the regular power average I got a lower SPL as you divide by the frequency.

Wait, what? Why would you ever divide by frequency? Frequency has nothing to do with this.

I think @pierre has it right - he's getting the same results as Klippel and his code follows the Klippel formula to the letter. (Whether Klippel themselves have the right formula is still up for debate, but given their reputation I would be inclined to trust them by default…)