[
Top of Page ]
[ Contact Us ]
[ Support/Wish List ]
[ Copyright © 2008, Audio Asylum, all rights reserved ]
[ General ] [ Speakers ] [ Tubes ] [ Vinyl ] [ Digital ] [ Hi-Rez ] [ Video Asylum ] [ Cables ] [ Tweaks/DIY ] [ Music ] [ Films ]
83.84.178.82
In Reply to: RE: Studio vs high end posted by morricab on January 31, 2008 at 04:54:53
I'm not replying anymore to your unsubstantiated rants. The moment you 1. get the facts right and 2. remain factual and leave your comments in the drawer, I'll resume communication.
Here's what Floyd Toole says in his 2006 AES paper about Olive's correlation between anechoic measurements and subjective ratings, if you want to discuss the matter you should address yourself to the author, not me. Sean Olive is participating in the diy-audio forums, or was it audioholics?7.1.1 Correlations between Subjective and Objective Domains
For over 20 years the author and his colleagues have conducted comprehensive anechoic measurements on many loudspeakers, and have examined the results of double-blind listening tests performed on those products. The results have been gratifyingly similar: loudspeakers exhibiting certain generally recognizable measured characteristics consistently achieved high scores in subjective evaluations [58]. As subjectively interpreted, a smooth, flat, wide-band axial frequency response, combined with similarly well-behaved off-axis responses, up to and including sound power, appeared to be the desirable pattern. What was missing was a mathematical process by which the technical data could be converted into a figure of merit, an estimate of a subjective rating of sound quality. In 2004 Olive took up the challenge and, axing subjective and objective data from 70 loudspeakers, developed a model to perform the conversion [61], [62].
Earlier attempts had been based on basic measurements such as sound power or room curves, usually with restricted (such as one-third octave) frequency resolution, Using more recent psychoacoustie knowledge, the new models examined much smaller details in the raw measurerments and, having access to more information, they could apply different weightings to the direct, early reflected, and late reflected sounds, When all of the now perspectives were included in the analysis, the result was a correlation of 0.86 between the calculated subjective rating and the real subjective rating. So it seems that we truly are measuring quantities that are important to our subjective tastes. It is not an accident.
As impressive as this is, it should be noted that there was a significant source of variation in the subjective data. The 70 loudspeakers were evaluated in 19 different listening tests, conducted over a period of nine months. In each test, only three or four products were compared, so, inevitably, there was a certain amount of drift and elasticity in the subjective scales used by the listeners. In other words, depending on what other products it is being compared to, and how long it has been since it had previously been auditioned, the rating of any individual loudspeaker could move up or down the rating scale. The movement is usually not large, but it is a change that the statistical analysis regards as uncertainty about the rating, reducing the correlation.
To overcome this, all loudspeakers must be evaluated in one continuous test, with each product being compared to every other product. When this was done with a group of 13 bookshelf loudspenkers, the correlation improved to 0,995—near perfection. The fact that the loudspeakers being compared were of similar physical configuration was an advantage, but that does not detract from the importance of the result. It is clear that there is a way to translate anechoic data from loudspeakers into very reasonable predictions of subjective ratings as they occur in a normal listening room.
And there is more. The excellent correlations mentioned came from a model that had access to a complete library of anechoic data—70 individual high—resolution frequency response curves at different angles surrounding the loudspeaker. With less data the correlations were less good. High—resolution data (1/20 octave) were consistently better than one—third—octave data. No single curve, anechoic or in-room, alone was adequate, although the axial response figured prominently in all of the successful models, perhaps because it is the event that triggers perceptual processes like the precedence effect, and how one perceives later arrivals. Early in this paper it was noted that reverberation is not a dominant factor in what we hear in small rooms, and here it is no surprise to find that the sound power output from a loudspeaker is, alone, an imperfect predictor of sound quality, especially when, as is commonly done, it is one—third—octave filtered.
Edits: 01/31/08Follow Ups:
Top of Page ]
[ Contact Us ]
[ Support/Wish List ]
[ Copyright © 2008, Audio Asylum, all rights reserved ]