Should I eq the speaker to anechoicly flat then put in room, or use some target curve you prvided from your link? The target curves seem to boost the bass, except for Dr Toole's curve. Is that bass boost equal flat anechoic speaker out in room, or flat anechoic with bass boost?
If you have the Listening Window (LW) and anechoic on- and off-axis curves of your speakers, you should use those as additional reference for applying speaker EQ. If you don't have those, you can use a moving microphone measurement(s). It doesn't have to be equalized completely flat as long as you are able to make it more linear and smooth. The amount of tilt on either end of the curve is up to you (also depends if it's an RMS avg, vector avg, MMM, single or summed -- though, I probably would try not to deviate too much from the natural response of the speakers in the room -- assuming the speakers are already linear enough (anechoically) to start with.
At 20 Hz one cycle is 50ms, 5 cycles will be 250ms. Is it really caused by too narrow gating? I am not experts in definition of different graphs, but I remember CSD is not reliable at low frequency because CSD is gated, what about wavelet spectrogram? Is there a graph that looks like CSD but without any gating?
You can change the parameters for most of the views that include a time component -- e.g. GD, decay, waterfall, spectrogram.
I'm no expert myself (and generally do not bother with the math), but some of these graphical views are rather intuitive enough -- it just takes some getting used to "reading" or interpreting them.
example Wavelet and freq magnitude graphs:
*Dips in the windowed response correspond to delayed energy arriving at the microphone position.
The two curves are derived from a single left and right average:
1.
Green is an RMS average (magnitude information only) where I
artificially generated a "minimum phase" response (essentially there is effectively
no time delay).
2.
Red trace is a vector average where the original time information from the impulse responses are included in the summed/averaged result.
"Artificially" generated minimum phase version wavelet of the RMS avg; and magnitude traces without any windowing:
The higher wavelet resolution (1/24) I used emphasizes the frequency magnitude component of the IR.
We see that some of the decrease between 150-400 Hz is caused by the (room induced) phase difference between the vector averaged, surround left and right (SLR) channels.
We can see some of this information already repeated in the overlayed group delay graph:
Right now, I really don't know what windowing works best... but it's kind of a case to case thing as well. I just cycle though different settings. However, in the low bass, I generally do not feel the need to apply any kind of smoothing or windowing at all.
*Also Attached the MDAT file for reference.