Google’s speakers have had local voice processing for basic commands for years now. Things like toggling power on lights or changing TV volume. They’re also expanding that locally processed control into then being run through the Matter standard so the whole chain is done locally. I have personally verified the local processing on my own network traffic.