@Colonel7 and
@Bear123
I'm pretty familiar with the JAES publications from the Harman group, including Toole and Olive. From what I recall, in Olive's AES convention paper (2004), his group presented a regression formula to predict listener preferences of loudspeakers based on loudspeaker measurements, which is what I believe you are suggesting can be done, albeit with high reliability. You are even willing to go as far as to say that you can predict that the B&W 800 Diamond loudspeakers "are highly likely to sound poor" based on measurements published by Stereophile.
I find this claim hard to believe. Consider the regression formula that Olive developed to predict loudspeaker preferences based on their measurements:
Preference Rating = 12.69 − NBD_ON * 2.49 − * 2.99 NBD_PIR − LFX * 4.31 + SM_PIR
NBD_ON = Average Narrow Band Deviation (dB) in each ½-octave band from 100 Hz- 12 kHz of the
ON-AXIS anechoic FR curve
NBD_PIR = Average Narrow Band Deviation (dB) in each ½-octave band from 100 Hz- 12 kHz of the
weighted average of the on-axis, early-reflected, and sound power measurements curves
LFX = Low frequency extension (Hz) based on -6dB frequency point transformed to log 10
SM_PIR = Smoothness (r^2) in amplitude response based on a linear regression line through 100 Hz -16 kHz of the
weighted average of the on-axis, early-reflected, and sound power measurement curves
What I couldn't help but notice was that, their paper, EVEN AFTER applying this analysis on 4 different measured curves for each speaker, some anechoic and some in-room (with and without a time window), the formula STILL could not reliably differentiate between the best loudspeakers. I mean, it was pretty good, but nothing to hang your hat on. Take a look at the chart below. For loudspeakers with the highest predicted scores (6-7 range), the perceived preferences were still scattered across the board between 5 and 8 (a very large relative range on their scale)! And there is considerable overlap between speaker preferences when comparing the top quartile with the 2nd quartile of measurement performance.
So, I guess my question is, if the Harman group can only "sort of" predict which loudspeakers will sound best after applying a complex formula which applies statistical analysis to a series of measured curves, why on earth would you think that you could simply "eyeball" a subset of those curves and somehow achieve greater predictive capability?
View attachment 68325