After a few hours of training. I realise that it is learning my reading voice instead of my natural voice, because to train the system I am required to read text. That means the system works fine as long as I speak in my reading voice.
This is a bloody pain, but it is still quicker than actually typing the text myself. So if I am writing a long reports I might consider using voice recognition when I can't be bothered typing or use it to cut down on the amount of typing I have to do.