training voice commands

Um, the car doesn't learn your voice. You have to learn how to make the car recognize your command. That's why most people find the Voice Recognition System primarily a source of ammusement. How else to explain "Display off" causing "Showing Bank Icons"?

Run through a bunch of commands and see which ones work for you and which don't. Do it first sitting still, then do it again when the car is moving and full of road noise. Figure it's a clever idea, then forget about it.

 

And you simply must try "I'm hungry."

 

As Bill said, you have to learn it. FWIW, I did find that personally I was talking too loud when trying to get it to work. When I just tried in a normal speaking voice, I had more success. However, I still get the "Not even close" responses.

For the most part I don't use it, however I still try for some. Some commands I can get to work relatively consistently are:
"Hungry", "Entire Route", "Single Map", "XX Degrees" and "Audio on/off"

Bob

 

I stopped saying Yes when it took four tries one day to get the car to respond properly, and since I had to hit a button to be allowed to say Yes (design flaw, in my opinion) I opted to press the Yes button that comes up, instead...

 

How unfortunate that such an advanced car has such a stone-age voice recognition system.

 

Actually, it's an advance in technology that's allowing these low-power computing devices to recognize a wider range of speaking styles more accurately without training. It's hardly stone-age, but it's not perfect, either. A system like this actually makes better sense in a car than does a trainable system because if the system depends on the training then there's a better chance that stray background noise will interfere with recognition.

The microphone can be directional (it is near the map lights) and a band-pass filter can cut out a lot of background noise, but in a car you're never going to be in a perfect situation. You could train with road noise or without... or train it twice... but it'd probably have the same accuracy as it does now.

Besides, this way no one has to worry about storing training profiles if there are multiple drivers or the vehicle is sold.

 

Good point.

It seems like the voice-recognition technology would be something that someone at Google is playing with during their free-time. Voice-recognition is kind of like searching and indexing the internet - it's just data in the end that you're searching for....hmm...what a great project.

I'm a software developer, so I tend to look at these types of problems in a programatic way. One of my "keys/rules of being a good programmer" is knowing how to massage the data and transform it to make your life easier. Most data problems that are difficult can be made a lot easier if you convert hte data into the right format. Perhaps we haven't found the right data transformaiton to make this an easy problem yet?