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

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.)