If you have been reading our speaker reviews, you have no doubt seen the "Preference Scores" for speakers. This was ground breaking research by Sean Olive published back in 2004 with the goal of predicting listener preference using anechoic chamber speaker measurements. Seemed like an impossible task but Sean pulled it off going beyond people's intuition that "everyone prefers a different sound." Clearly if we can predict preference based on measurements, then it is listener independent.
...
Conclusions
There are none as yet. I expect this to be a living research thread where we discuss what we have found here, and whether we can better rationalize speaker preference from measurements. The preference score for this speaker will be high (
@MZKM will post shortly) putting me once again at odds with it. We have to figure out why before I lose all face.
Amir, I had posted the following quote on AVS Forum last year but hope you won't mind my reposting here. Although I was comparing the curves of the F228Be vs the Salon 2, and none of this is new, I think much of it is relevant here.
"1. The curved baffles on the Salon 2 reduce diffraction, which I wondered might at least partially explain the MUCH more similar on-axis and listening window responses for the Salon 2, as opposed to the F228Be that shows more differences between 700-800 Hz and 1-10 kHz between the two response curves.
2. Both of the Olive models use smoothness and/or flatness of on-axis but NOT listening window response as dominant factors. The on-axis smoothness (narrow band or not) actually outweighs the bass extension in both models, so whether the bass or the on-axis response gives "the edge" could be a matter of contention
3. Although the F228Be initially appears smoother overall, the two significant deviations in the on-axis curves are bumps around 700-800 Hz (broader) and 5 kHz (narrower). These are of similar magnitude and width in the sound power curve, which is also a direct factor in one of the Olive models and indirectly (though the predicted in-room response) in the other. The Salon 2 bumps are generally of less magnitude in the on-axis curves, also less magnitude and/or width in the sound power curves, so one tends to smooth out compared with the other. I didn't know whether the F228Be bumps could represent some sort of resonance, but these could adversely affect its audible performance until equalized
4. Dips in the on-axis or sound power curves are less audible compared with bumps.
5. The Salon 2 does have a larger dip in the sound power curve (above 2 kHz, almost reminiscent of the BBC dip but likely due to the crossover and directivity of the drivers), but the sound power curve is a significantly smaller factor in both Olive models compared with the on-axis curve, plus less audible as above in 4
6. Wider bumps or dips are more audible than narrow ones, but dips are less audible than bumps (#4)
7. The F228Be's 700-800 Hz bump is relatively wider and higher than any of the Salon 2's, using a broader baseline, like 500 Hz-10 kHz
TL;DR The F228Be may look smoother but may not be as smooth when and where it counts, may have flaws of commission that are more audible rather than omission, and may have less bass?"
Additionally, when I asked Kevin Voecks about some of these points, he responded "I suspect one of the differences is heard when they are played louder, where nonlinearities can occur. It is especially at moderately high levels where I hear the biggest difference. (I don't play anything dangerously loud.) The other possibility is as you stated, the almost total lack of diffraction with the Salon2s."
It's clear that some of the issues above apply when comparing the IL10 and M16, particularly when looking at the resonances and the PIR curves.
Others have already pointed out that the speaker preference models were developed using speakers of extremely varying quality, so it's reasonable to expect predictive issues when applied to a group of speakers of relatively high quality. I vaguely suspect that an updated speaker preference model would differentially weight Deviation (whether AAD or NBD) for bumps versus dips, as well as take into account the Q for bumps, plus factor in nonlinearities like distortion and perhaps effects of diffraction, which I think is reflected in the relative similarity between on-axis and listening curves.
Young-Ho