Re: you could simply not put the creepy things in your home
I find it hard to believe that a current fast desktop or laptop machine can't handle speech-to-text
I imagine the majority of the computing power is required by the text-to-request phase of command interpretation. Text is really an approximate form of phonetic encoding, so its relationship to speech is more-or-less finite. Extracting the meaning from the encoded request is a more open-ended problem.