I agree, the voice speed is too slow, and it sends the data back to google for analysis each time, rather than doing it on the phone, which means if you are anywhere with less than stellar reception, or if you have limited data, it can become instantly useless. If you could download the maps for you local area, and do vocal direction searches on it, or search your phone without having to be connected to the internet, I could see myself using it a lot more. As it is, I have found that as useful as voice recognition might be, in reality it is far out-shadowed in usefulness by google gestures, which is quite simply the greatest phone app I have yet discovered. I have gestures set as one of my four taskbar apps in Cyanogen, and in the quickbar, this means that from anywhere In the phone i am only ever two touches away from it.
What google gestures does, for those who have not tried it, is to index everything in your phone, contacts, apps, songs, games, the lot. it then allows you to search through it all by drawing a full screen letter. If a single letter does not narrow the search down enough, wait a second and enter in a second letter. I have never had to enter more than three letters to have the correct item appear, backspace is back swipe, clear is forward swipe. All processing is done on the phone, results are instantaneous and every result gives the option to play, view, edit, send message, call, locate or delete right from the same window. It is genius.
I never seem to see it mentioned in android reviews, but for me this is androids killer app. Combined with swype or shapewriter it totally changed the way i interacted with my phone. Voice recognition may never reach %100, and it may require generalist AI before it does, but contextual gesture recognition is here now and it works fantastically. While iphone has some apps that perform similar task, they are all for a specific purpose, like swype but just for sending texts. Whereas android allows you to set swype as a system wide input default.
Ubiquitous seamless context sensitivity is just so beautiful, it is hard to imagine it not becoming a yardstick against which things are measured. Instead of asking if can email or skype, we may well be asking if it can detect when I'm at the movies and turn itself on silent, or adjust its brightness to match the setting, or make the heater turn on when im getting home. So much of what we use our phones for is predictable, as the sample set of your daily routine grows, your devices should be able to learn and adapt, just as its dictionary builds a lexicon of your used words, so it should build of memory of your default actions and be ready to activate proactively rather than reactively. Why should I have to hunt for the TV remote app, it knows that i am infront of the TV, and that the TV app is what I usually use when I am there, with a bit of context sensitivity the whole debate around ease of use could rightly shift from the sleek ui, to the sleek life integration, which is really where google shines.
Oh and in regards to your issue of going back and editing, I had the same issue, until I finally discovered what that seemingly superfluous trackball was for, precise letter by letter control.