Originally Posted by mogorf
This is wrong information. The Audyssey algorithm does not give priority to "problems" found only at one mic spot. It has a pattern recognition system, clusters the mic measurements, looks for similarities and weights them with fuzzy logic to create the final filter. Right?
No it is not wrong. Explain how electronic EQ can correct a 7dB peak at seat number 4 without affecting the same frequency heard at seat number 2.
Say the peak is at 400Hz. Audyssey measures using all its resources and it detects a 7dB peak at 400Hz at one point in the room, which we are calling seat number 4 (this is arbitray - it's just where the mic happens to be placed when it detects the peak). Audyssey then creates a filter which pulls down the peak. However, there was no peak at 400hz at seat number 2, but the frequency response has changed for ALL seats and so seat number 2 now has 400Hz pulled down by 7dB even though there was no 400Hz peak at that seat. Obviously, Audyssey cannot create different frequency responses for different seats, at the same time.
No matter what "fuzzy logic" it uses, it cannot do the impossible. So the result has to be some sort of compromise, where neither seat 4 nor seat 2 is corrected properly but both have some
correction, making one seat slightly better and one seat slightly worse.
Clearly, Audyssey does not use some simple averaging. If it did then it would encounter another problem: if seat 1 had a 6dB boost at 300Hz and seat 5 had a 6dB dip at 300Hz, then simple averaging of those two positions would mean Audyssey would see no problem at all and do nothing. That would be a ludicrous solution.
More likely, and nobody knows as this is proprietary information, what Audyssey does is this: it measures the response at various positions and it looks for similarities and differences. If it sees the same, or essentially similar, problem at all seats, then it knows it can safely fix this problem, because all seats will benefit. If it sees just one seat has a problem, then it can ignore that one seat and assume the problem is an outlier - creating the greatest good for the greatest number. Or, it can attempt some form of correction of that problem, but if it does, whatever correction it applies will also apply to all other seats (as EQ cannot create different frequency responses for different listeners/seats/positions).
This is why the whole 'religion' that correcting for multiple seats is the only possible way to go, is totally wrong. If one is solely concerned with one seat, as many are, then correcting for that one seat resolves all of the above problems at a stroke. However, if one wishes to give a good result for a larger number of listeners, then inevitably there has to be compromise and each seat will get a good result but no seat will get an optimum
result. So in the end, as I said, it is a matter of choice depending on circumstances as to which route one wishes to take, with neither being 'right' and neither being 'wrong'.
Audyssey XT32 is pretty good, but it cannot do the impossible