Voice recognition software today is very good, but when the computer is trying to match your voice to over one million possible things that you could have said, it fails some of the time. Plus as humans we say a lot that we don't mean and controlling your voice while recognition software is listening can be a little awkward at times as you want to be expressive with your voice, but can't be because the text will read back wrong in the mind of the reader.
I don't know if this problem will be fixed in my lifetime as the solution it isn't simply an increase in computational power or better quality microphones. Computers will have to learn how to listen to a human and then rewrite what they have been told to write on the screen.
Human> "and this will result in a less errors replicated on screen"
Computer Types> and this results in fewer errors being typed by the computer in documents
