Monday, April 7, 2014

Arguing with your car

My previous car - a VW Tiguan - had a certain amount of "voice recognition" that could be used to attempt to control certain functions hands-free. Mostly it was for the phone interface. It was mildly successful, by which I mean it caught about 30% of what I was attempting to tell it. Didn't seem to matter if I was driving or not - ambient noise could be high or low - I always ended up arguing with my car. Invariably it would call the wrong person, or when I accepted the option it chose, it would cancel the call (because in VW-speak, apparently "yes" sounds just like "cancel" despite sharing no similar sounds or even having the same number of syllables). So when I bought my latest vehicle and they asked if I wanted "voice recognition" turned on or not (it's a dealer-activated option via the OBD-II port rather than a driver-activated option), I told them "no". The guy was really confused by this, and went on to explain to me how awesome it was that I could control everything from the radio and phone to the heating controls. For shits and giggles, I had him turn it on and after ten minutes of not getting the car to understand a single command, I had him turn it off again.
Here's the problem : nothing I own that has "voice recognition" can understand me with any level of reliability. Not my car, not Siri on my phone - or Google for that matter. Not my Xbox One, not Dragon Dictate (or Naturally Speaking). Hell - even the phone systems where you have to speak to them can't understand me (Delta flight tracking 1-800 number, I'm looking at you). Like my VW, they all get about 30% of what I'm saying, and the success rate goes up a little if I switch the language to Australian instead of English, or if I fake an awfully stereotyped American accent. But other than that it's way quicker for me to just use the normal buttons, knobs, switches and other methods of input. As in I can tap "send" to send a message or email instead of arguing with my device over whether it heard "sand" or "bend" or "vend".
Because I have so many of these technologies that can't understand me, I have to conclude that voice recognition is either just crap in general (and everyone is so blinded by a 30% success rate that they think it's marvellous), or nobody has managed to build one that can understand an unaccented English voice. It's not just me, either. My wife and all but two of my friends have the exact same problems. We spend ages arguing with "voice recognition" systems when it's quicker and more accurate to just press the damn button yourself.
The problem with all of this, when it comes to cars, is that you can't rely on it. Why have "voice recognition" if the driver has to double-check every action? It's quicker and safer to just do it yourself once, than it is to argue with the car for a minute, check what it's done, then hand-correct it. Same for phones or dictation systems. If it doesn't work 100% of the time, you can't rely on it, and once you get into the mode of having to double-check everything, you realise it really IS quicker to circumvent voice recognition altogether and just do it manually. This is a problem when it comes to texting and driving. I would love to be able to get my phone to read texts to me and take replies while I'm driving. But it can't. "I'm leaving work and I'll be home in about 20 minutes" comes out as something like "I'm breathing burp and won't be on around plenty bins". Thus the temptation is to text while driving, which is a distraction, which can lead to accidents. So I don't bother.

Oddly enough, the only phrase that Siri ever gets 100% of the time is this: "Why don't you understand English". To which the response is a predicable "I don't understand why don't you understand English". Try it.