Speak for 11 seconds otherwise Skynet will know that you are inebriated. Confused? Read THIS, then come back.
Now, my first inclination is to think that the creators of this speech recognition software are tooting their horn a little bit too much. I am thinking of early lie detector tests where foolproof claims were made that just could not be backed up. Primarily because body reactions may not be about what you are being asked but what someone is afraid will be learned.
A person could have just gotten a yes to a date proposal, for example, and be really giddy. Maybe as giddy as I get when tipsy. Or they could have been told no and start mumbling. Would that trigger as an alcohol caused speech pattern.
And going back to my happier mood, how does the AI know my baseline? And what if I am particularly happy because my birthday month starts tomorrow? How does that factor in?
As with the lie detector, this inebriation sensor might be good as part of an overall set of proofs, not as a sole one.