At one point Google had highly specialized hardware that only listened for “ok Google”; that’s why you couldn’t (and AFAICT still can’t) change the wake word.

Things may have changed in the years since I learned that, but I suspect recognizing a bunch of words from an ever-changing list would still need to be done in software and require the phone’s CPU to run.

OTOH, the way Android phones recognize and songs for you is very much like what you described, so maybe there really is hardware already that can recognize a shitload of arbitrary sounds using practically no power.