As a court reporter I use a combination of speech recognition (Dragon Naturally Speaking 11.1) and my home-brewed stenotype shorthand with WordPerfect for Windows and a Maltron dual hand 3D keyboard. My first version of Dragon was 3.0, and it's steadily improved since then.
It's easy (in fact almost essential) to use the two systems in a complementary fashion. I don't type much from written sources, but using practice pages from an old typing manual I've achieved over 210 wpm with the Dragon at an accuracy of about 98%. One of the necessities of using the Dragon is that there must be a pause before speaking a command to the computer. The pause is user-adjustable but it cannot be too short or it could result in the computer misunderstanding what is said, and executing a command instead of recognising as text. Thus I've found in a situation where there are likely to be a few commands (particularly telling the computer to "press" a certain key) intermixed with text recognition, it is actually more time efficient to use the keyboard directly rather than the speech-recognition.
Whilst it may sound easy, (as mentioned above) using a speech recognition package is actually a fairly technically taxing undertaking, and doing it incorrectly is likely to result degradation of the system's performance over time.
Joe