Tone and inflection. Current voice recognition systems (of this type) require a command word to clue in that it should be accepting commands. Future voice recognition systems will rely on tone and inflection. The same way when you talk to a baby in a "baby voice" it understands you are speaking to it. Or when you're talking to a dog, other humans around you, even blind humans, will not mistake it for a conversation! Some day computers will do the same, and humans will adopt a "command mode" voice for interfacing with computers. I read some early prototypes are using "whispering" voices to indicate command mode.
Calvin