It's not so much the "fingerprinting" - it's more that there are some words ("type 1") that have many relatively frequent words that sound like them, and there are other words ("type 2") that only have a very few other words (and infrequent ones at that) that sound like them.

Type 2 words make better potential wake words.

(That's not the whole story, but I think it's good model.)

from my observations of the masses that Alexa and co try to cater to, that's too much work or too technical. we have to remember the average joe/jane don't know what a browser is.