Doesn't alexa (or smart speaker) always need network to transfer voice data to server ?

I doubt they have ever ran machine learning model on those speakers. Mostly just some simple one like "hey alexa"