If you want to have fun: put sound-muffling headphones on someone, record their voice, pitch-shift it by half an octave and feed it back to them live.
The processing lag of the recording and processing is usually a few hundred milliseconds, which is enough to mess with people’s speech on its own. But the pitch shift should send their voice completely loopy, as their brain tries and fails to “correct” the pitch, swinging to alternating extremes.
There’s individual variation in susceptibility, and the effect is much stronger on self-generated speech rather than read text. It’s quite dramatic when it works, though.