It seems to me that voice recognition is not the most efficient way to interact with a computer, especially when the user interface is well designed. For complicated tasks, and for interacting with computers where you may not have a normal desk or terminal, perhaps. As far as voice-to-text, if the recognition is accurate, it can possibly increase productivity depending on the person and their typing skills. On another not, however, this is a way for paralyzed individuals to interact with computers without the use of traditional means.
However, using voice interaction in tandem with other means could be a more efficient route. Having a computer run commands in the background via voice commands while you interact with it in more traditional ways in the foreground.