How many times does the person say shout this? If it's just a few I would split it on both sides to create a new item and click on it to bring up the item properties, and then lower it enough so that it fits in the general dynamic range of the rest and still retains some intensity. So that it sounds nearly good enough before using compression.
The other advice would be to not use a single compressor to tackle this issue. Use two, the second being set more like a limiter. Try this: on the first Recomp have a low ratio, around 3:1, attack 20ms, release 70ms, knee around 5. Set the threshold so that it just knocks down the program around 3 or 4 db tops. Then have it followed by a second instance set more like a limiter. ratio 30:1, with the attack and release a bit faster, 15ms and 60ms, knee at 0. Set the threshold so that it ONLY kicks in on the transients that jump through the first Reacomp, not all the time. You can adjust this so it works best on the "Yeow"s and then check how it works on the rest of the program, or find a good threshold on the program, where it's only reducing the top transients and not kicking in otherwise and check that setting on the "Yeow"s.
Reacomp isn't really exactly being a limiter here but it's being limiter like. There are several good limiters in the JS fx section (mine has plugins from different origins, mostly arriving via Repack, so I'm not sure which come stock with Reaper, but anything that's named limiter you should give a whirl). As a simple tool for podcast voices you can't go wrong using the free Loudmax vst plugin as the limiter after the compressor
https://loudmax.blogspot.com Set the output desired and the threshold so that it only knocks down the transients that make it past the compressor.
If you have a bunch of instances of the shout it might be worth it to tame it with plugins. If it's only there two or three times just use the volume envelope or item volume to bring it down slightly, to be more manageable but not lose intensity, and then have compressor followed by a limiter do the rest. It's entirely possible you won't be able to have both the level you want and not lose more intensity than you want no matter how you do it. The sound of a loud shout that's brought down to match the level of soft conversation is likely going to be weird or unnatural sounding. That's just the nature of the beast : ) You do want to protect the ears of the listener and compromise the intensity of the shout if you have to.