My program implemented Voicebrook around the time I was leaving. Frankly, my experience was pretty positive. I had had bad experiences with voice recognition previously, but the recognition was not bad at all. Without having any specific data, I don't feel like I spent more time editing gross descriptions from Voicebrook as compared to our human transcriptionists of yore. Even medical terminology seemed to get picked up pretty accurately and the "correct this" or "spell it out" features were decent. Thus by some metrics the program works very well.
There are serious caveats. The interface with the lab information system was clunky and this is the system's Achilles heel. Voiceover is an overlay on your preexisting LIS and relies heavily on having the right window in the foreground, having the system not time out during long dictations, etc. As soon as you do something unconventional (even clicking in the wrong place) you may lose your dictation. Finally I did not get the impression that the Voicebrook team really understood pathology and they basically denied all shortcomings in their system.
