AFAIK it isn't actually possible. This is from memory, but there's technical limitations to trying to listen for and process everything all the time, so there's a secondary processing unit whose only job is to listen for the activation phrase and activate the main unit when it hears it. I believe this was hardwired to only listen for the phrase itself and doesn't use the same speech recognition model

Devices like this aren't hardwired into anything. Echo is sold in many markets with different pronounciation. Amazon isn't going to scrap thousands of units when the Spanish market doesn't want them but the US does :) Or when they introduce a new market.

Indeed the hotword is listened for by a local CPU, not the remote cloud which interprets the rest of the sentence received. But you can bet this is not hardwired in any way.

These devices are field programmable in full. Just nobody bothered to hack them for a measly 1000 bucks :P