Cousin Jon e-mailed me a New York Times article on the latest version of desktop speech recognition using Dragon technology.
We have a sort of familial interest in speech recognition, since both of us worked for Dialog Systems in the 1970s.
By the time I had left, Dialog had developed speech recognition systems that worked over phone lines and recognized 'isolated' words. In other words, the systems did tolerably well as - long - as - you - paused - between - each - word. There was also an R&D demo system (paid for via Rome Air Development Center) that listened to continuous speech and identified specific words (perfect for monitoring phone conversations for words like "terrorism" or, back then, "Russian spy")