While I was playing with word highlights I tried something more extreme as a side experiment: a separate macOS window that shows only the sentence you're currently hearing, nothing else. You toggle it from the reader toolbar and it floats there, one line, in sync with playback.
The words don't just appear and disappear. They grain in with a Metal noise-dissolve when the sentence arrives, like text developing out of static, and then dissolve away one by one as the reading cursor passes them, so the sentence slowly erases itself behind the voice. The current word is drawn in the accent orange, and the color eases as the cursor advances, so the word being spoken still reads at a glance.
Getting the dissolve to feel right took a few tries, and most of the fiddly part was the last word. I first eroded each word gradually as it was spoken, driven by the same sub-word progress the karaoke fill uses. It looked nice, but the final word had no next word to trigger its exit, so it blinked out when the view swapped to the next sentence. I went back to a cleaner all-at-once dissolve when the cursor passes a word, and made the last word vanish once its own reading is essentially done instead. It's macOS-only and firmly in the experiment column, but watching a sentence erase itself as it's read turns out to be a strangely calming way to follow along.