That (tonality) is not what you are asked to score. You are asked to score your overall preference for one loudspeaker versus another. Again, this is what we all do day in and day out. We listen and compare and give an opinion of what we like better. You are free to use whatever metric you want in giving that final score. Use emotion, logic, analysis, whatever. You are not given any bounds just like you do in sighted evaluation...
...I am stating what the experience is, not the rules of the test. The rules of the tests are exactly as I mentioned: score them loudspeaker preference from 1 to 10. That is it. Now, sit in the chair and listen and the first thing that hits you in the face is how different tonally each loudspeaker is. I am sure everyone agrees that every loudspeaker has a different tone to it. So it is natural to judge that, given how large and obvious the difference is. High correlation to frequency response measurements shows that this is indeed what people naturally do, without being instructed one way or the other.
But again the tests are pure preference with no restrictions.
Okay, just so I’m clear on this:
Harman subjects its speakers to 70 individual frequency-response measurements at 10 degree intervals in an anechoic chamber and then conducts blind listening test for preference where…
1) The room is of no resemblance to the original space the measurements were taken in
2) The measurements use steady-state signals but the blind testing uses music
3) The speakers are not matched with regard to driver composition, number of drivers, crossover frequency, crossover slopes, nominal impedence, minimum magnitude of impedence, electrical phase angle, cabinet resonances, step response, cabinet construction, etc
4) The subjects are not screened for conflict of interest/hearing irregularities/ability to correctly identify tonal variations under ABX testing
5) The subjects are asked to give an arbitrary numerical rating in which there is no consistency of metric in application
…and then claims
high correlation between smooth on/off-axis frequency response and preference where NONE of the above variables are eliminated?
I… I…
If I am testing for preference of a specific variable (frequency response) and want to claim a statistically significant correlation between that variable and preference,
but do not eliminate or control all the potential confounding variables that may adversely affect the relationship between the independent variable and dependent variables, I will damage the internal validity of the experiment. In short, I will not have correlation, I will have an anecdote and I will fail Psyc 101.
Do you really feel completely comfortable calling the above testing protocol “audio science”?
You mean if someone presents same or similar spatial averaging loudspeakers to me in a sighted evaluation like we all do, I can't rely on my ears? I will be relying on other senses alone like skin conductance? Can you give an example of two different loudspeakers where my ear would say they sound the same but these other senses say different?
No, you could rely on your ears, but if we were testing two different speakers that sounded similar, and the subject had difficulty articulating the differences, skin conductance response, heart rate, respiratory frequency, skin temperature and facial EMG would perhaps tell us whether these physiological responses were indicative of neurological processes at a sub-conscious level the subject is not aware of.
Remember, the #1 factor in sales of loudspeakers is marketing. Nothing remotely approaches the power of that. So let's not use commercial success in this context. It belies the reality of how the market works...
...Many loudspeaker desigenrs rely on this research in producing their loudspeakers. Their value therefore is established, lack of awareness of high-end consumers notwithstanding.
Which is tantamount to saying even the flakiest of measurements made in the name of science are of value for the purpose of marketing. I agree.