You shouldn't have to press a button to activate speech recognition. It should have pretty robust recognition of a code word. Versions of the technology are used for word spotting in which the system "listens" for specific words or phrases, often in noisy channels. IIRC, TI was the main player in developing word spotting. Most of the cites supported by DARPA concentrated on speaker independent recognition of continuous speech, which transitioned into spoken language understanding systems that incorporate language modeling and expert systems.
db
db