Re: contour EQ
Philip sent up the bat signal via facebook ;-)
Wow - what a thread! This is one for the archives.
Now - how to keep this short? Not possible. There are too many topics worth commenting on.
Smaart
The key to understanding Smaart's "hybrid" windowing is its use of a continuous signal (typically pink noise) as the stimulus.
Consider a Smaart setup with the offset delay properly set. The first data point in the measurement record is entirely a response to stimuli that happened before the beginning of the reference record; so any correlation with the stimulus is accidental. Now consider the last point in the stimulus record: the entire response to it will happen after the end of the measurement record; once again, no correlation.
However, if you consider the middle point in the measurement record, the stimuli that produced it are mostly captured in the reference record; so nearly all of the direct response will be highly correlated. So, the further in time a response is from the stimulus that produced it, the lower the correlation of the two: the early part of the impulse response is highly correlated and the late part of the impulse response is progressively less correlated.
In essence, the Smaart display combines a windowed direct response with an RTA of the reverberant field. The relative weighting of the two depends on the length of the record and the direct-to-reverberant ratio. At low frequencies, where there is a lot of reverberation, the room has a correspondingly larger effect on the result; whereas at high frequencies, the direct response dominates. That's one reason why the displayed HF level normally increases after you set the offset delay.
... which provides a nice segue into my tuning comments:
Tuning
If you put a flat speaker into a reverberant room, its measured response out in the room (using Smaart) will not be flat. Because reverb generally falls with increasing frequency, the measured response will usually ramp up toward the low frequencies - just like the curves posted earlier in the thread. If you shelve the low frequencies down to make the curve flat, you don't just reduce the reverberation, you reduce the direct sound in that range; resulting in a system that sounds harsh; low-mid heavy instruments (especially transient ones) that sound small instead of powerful; and a midrange that sounds disconnected from the subs.
So my philosophy in that range is to let the system/room keep most of the room's overall character, and mostly use EQ to mitigate the effect of problem frequencies. Outdoors, there is of course very little reverberation; so the natural response of a flat system is pretty flat. A little low-mid emphasis (shelving up below 1000 Hz) will sound better on recorded music, make individual instruments sound a little bigger, and reduce the negative effect of having the highs blown around in the wind.
The appropriate amount of subwoofer boost depends on the style of music. Some sources like upright bass, piano, and bass vocal need their 80 Hz to be balanced with their 200 Hz to sound right. At the other extreme: in house music and most of hip hop, what happens in the subwoofer range is essentially a separate event from what is happening in the mains; so the subs can be radically louder than the mains and the two don't even have to be in the same place for it to work.
Esoteric Transfer Function Stuff
The reason for using cross-correlation and power spectral densities to obtain the transfer function, rather than just dividing the measurement by the reference, is that they're also the inputs for the coherence function. If you keep running averages of all three, you can get both the transfer function and the coherence function from the same set of averages. I'm not aware of any noise advantage to this approach, so I'd be interested to read the paper Nick referred to about that.
Cheers,
David Gunness
