You mix the dialog louder for the entire piece, and every other sound in the middle. No extreme highs, no extreme lows. General compression with dialog forward choices.
I don't know why compression isn't built in to consumer media devices, it's so often called for (and closely followed by volume normalisation ... but I guess the advertisers veto that).
> I don't know why compression isn't built in to consumer media devices
As always, it depends on the device. Dynamic range compression seems to be a relatively common feature, usually as an option described (inaccurately) with something like "Reduce Loud Sounds" like it is on the Apple TV.