How is the "ok google" command implemented?

Question

Google

How is the "ok google" function implemented? It would be more correct to say how the smartphone determines that this particular phrase was said? Does he scan the entire broadcast, record it and send it in chunks to the Google servers? And with the "Activate voice search on lock screen" function turned on, will the smartphone listen to everything that happens around it, recognize all speech, waiting for a command? That's tons of traffic.
The question is of interest because I wanted to create a "Smart Home" system with voice control with activation by a specific command, like "ok google". But he stood in a stupor, because it is irrational to recognize the entire ether.
Can someone point out a rough direction in which to dig.

Answer 1

Well, it is logical that the stream of sounds is analyzed, in the specific example "ok google".
And it is analyzed constantly, in the background, and without traffic, but by the internal means of the Googlephone.
This is an example of how a blocked phone receives information that an external call is coming to it.
So here too.
And only after the key "ok google" has worked - the phone tries to "hear" and recognize everything that will be said in a certain period of time or before a certain pause.