I tried this again on a new microphone. The Takstar CM-63. This microphone seems fairly forgiving of the spark interference in this process.
https://www.amazon.com/TAKSTAR-Instrument-Microphone-Professional-Gold-Plated/dp/B0C58XZ9RS?th=1
This was with the window AC running, since Texas is in deep fry mode. It wasn't at full blast since it was night and it's an inverter AC. This was the result of 100 sparks averaged together.
Controlling the interference is a huge issue, at least with my consumer grade equipment grafted together. The EMP causes an electrical impulse before the acoustic doublet wave hits, so the EMP becomes a step in the final impulse when the lowpass filter is applied. If the EMP is strong enough to cause demodulation then the step will become a sawtooth. This affects the midrange accuracy of the results.
My results are consistently droopy at 10k and above, compared to what the datasheets usually say. One reason for this could be electronic overload, compressing and widening the impulse, but moving the spark away does not always help. Assuming that some null in the output corresponds to a bessel function null does not seem to produce results that are any better. low-energy igniter type sparks seem to be a different beast with a different set of rules. But I don't have a Gefell mic to know for sure.
Averaging improvement of lower BW follows square root addition of noise, so if I get the lower frequency limit down to 1KHz with 100 sparks, to get to 500Hz I need 400 sparks. 250Hz, 1600 sparks. If the rig holds up for that long, it's doable. However room reflections become the limiting factor.
Here are various trial runs.