OK, so howto?

When I said the detection should be easy I meant the conditions are well defined. It should be easy for someone who knows how to do it but, as I say, I am a complete noob in software audio processing. Linux or otherwise.

I was even thinking about two daemons, one continuously recording audio in, let's say, one-second chunks, the other one picking up those chunks, analyzing them and deleting if found to contain silence. Definitely a battery - and possibly eMMC - killer but I only want to run it for one day and on a charger. (And stream to a disposable SD card.)

BTW, where did the tag come from?