I was astonished by the 3 dimensionality of my long time favorite Crime of the Century by Supertramp. But also thinking the voices slightly disembodied, as if the midrange were sucked out, but you were still hearing the raspiness.
My previously very ad hoc midrange EQ settings, designed to correct some apparent Acoustat peakiness, and yield a "Linkwitz-Gundry" dip, which I thought was pleasing, were:
2729 Hz, 1/3 octave, -2dB
5200 Hz, 1/2 octave, -4.5dB
9038 Hz, 1 octave, -2.5dB
To see how the voices would be affected, I turned all 3 PEQ's off. The result, somewhat counterintuitively, was less raspiness, and more wholesome sounding voices.
I had actually added the upper PEQ's to combat this kind of problem. But strangely they weren't helping.
It could be that the factors which lead to some peakiness in the mid highs are not minimum phase, and trying to correct them with EQ messes up, rather than improves, the phase.
I don't actually know what causes the apparent peakiness that I was trying to correct. I was not following my own guidelines of "correcting by the model not the measurements." I was just correcting to measurements, and mostly to sometimes highly misleading 1/6 octave RTA.
I'm not going to assume that "no eq" is the way to go. It simply needs to be done more carefully, including with tests like there where ALL of the PEQ's are turned off, rather than just one or two PEQ's.
For now, I'm back to scratch, with no midrange PEQ's (except the phase corrected crossovers at 125 Hz and 17000 kHz).
Later I was rediscovering Peter Gabriel. His voice sounded real but just a bit husky. I bet that's how it was recorded.
I need to set up my ABX to switch PEQ's using midi (as I have done before). The amplifier test setup is currently dysfunctional anyway (one DAC died...and I also need to get it fixed). And I have shown over and over that I can't tell the differences between Krell, Hafler, and Aragon amps, though I want to retain the capability of doing that kind of testing.
I also have a weird idea that introducing a small peak around 1kHz might even help. The original Rogers LS3/5a have a driver with a >10dB peak at 1kHz, which much of the crossover is designed to correct, but perhaps not perfectly. Perhaps the thing to do is to have a 1kHz peak with an equal and opposite 1kHz null. What would this do? Reduce phase information. That might actually make it sound better.
Vinyl records also have an inherent peak around 1.5kHz I believe, a ubiquitous cutter resonance.
But perhaps also the Acoustats have that very sort of resonance also, that actually compensates for the overdamped nature of the system, and trying to correct it to flat isn't right.
I should also try running Acoustats full range w/o subs and supers. I have the capability of doing that, I think, with a minute of knob turning.
Another weird thing is that from my adjacent room (kitchen) listening position, where I most commonly listen in background, voices and such inherently do sound more wholesome. It's mainly as you get right up to the Acoustats that they can sound a bit phase-y.
At the same time, a very close listening position can be astonishingly good, and improves the bass somewhat.
Swapping my high back chair for a low back chair did not affect the voice quality much, and though it makes a big difference in measurements it hardly seems worth changing for most listening. But still worthy of more testing.
It occurs to me that one way of improving the low frequency room modes is to have a FIR filter designed to move the effective origin of the subwoofer to the listening position. This can be done by creating a model of what the sound is at the listening position from the subwoofer at it's current position, then inverting it. This wouldn't yield "perfect" response, but the best available response without making other locations worse.
I think that same sort of "best available performance" is the secret of the Trans Nova amplifier design of the Hafler 9300. Feedback does not even try to correct the power supply sagging, which is maintained by the inherently high transconductance of the mosfets themselves. Feedback is taken from the gates not the output. Despite that, every measurable parameter (except max power), even high frequency high power (20kHz, 100W) damping factor, was equal to if not better than the Krell FPB 300, which uses feedback in the whole output stage by itself (hence very wide bandwidth...equivalent to the Hafler). When feedback is taken from the output, the feedback tries to squeeze more and more power from the power supply making it sag more and more.
Update May 8, 2022
Looking for an LP to play, I discovered an as yet unplayed Extended Version of Purple Rain by Prince. It sounded way way way too aggressive in the highs. Also very thin as if no midrange.
So I tried playing the same thing from Qobuz in high resolution. I sounded a bit better, but still too aggressive.
So I dialed in the old PEQ's, just as I had them before.
That fixed it. It's a very hot sounding album, but still listenable with the PEQ's I had dialed in previously.
It could be the PEQ's are fully perfected yet, but it appears that having them all on is better than all off.
