Unfortunately, due to the variations of spoken language, intonation, inflection, stress, colds, slangs and even accents will take their toll on speech recognition programs no matter how expensive the microphone or whether one is using Dragon 19 or whatever version.
The technology in itself is quite impressive but to obtain 100% accuracy on a routine basis is out of reach for all intents and purposes. Perhaps within a few years they will incorporate more AI into their speech rec programs that will allow the program to quickly infer and match what the user says and display the word(s) in proper context (meaning, usage, spelling, etc.).
I believe it's coming...just not quite ready for prime time as yet.