What kind of audio processing technology do they have here?
If I'm in a room so noisy the person talking to me can't understand what I'm saying, I find that there's no way in hell traditional, existing, speech-to-text technology will understand it. I mean, yes, the computer can go through the various things sound pattern 'X' might have been and deduce that I must have said one particular thing, but that's generally only useful in highly-scripted situations (the only recognized commands being "Balance", "Transfer", or "LET ME TALK TO THE ********** OPERATOR PLEASE").
I mean, really, hasn't anyone else used voice-command software in the past ten years? It still sucks.
Note that Nuance is a key provider of corporate and small-device voice recognition software. If you think your car's sat-nav sucks, you are getting a glimpse at how well Nuance's tech works in a constrained device.
That all having been said, it has got to be a huge improvement over Apple's current voice-dialing tech, but I just can't see any other situation where voice commands would be the primary way I'd want to deal with my phone.
I think you've got the direction backwards here.
If you're in a noisy environment, you turn on speech-to-text so that the voice of the person on the *other* end of the phone call has their voice converted to text. You then text back and your text is converted to speech for them.