Re: Does anyone know what the patents actually are?
Given the number of computers that have voice commands on SciFi films, doesn't this lot fail prior art or the obvious to practioner tests?
The patent in question is no doubt for a specific algorithm for speech recognition or voice-to-text. (It's not that uncommon to patent natural-language-processing algorithms in the US; IBM patented their Latent Semantic Analysis technique, for example, though frankly I don't know that many people wanted to use it anyway.)
Obviously there's prior art for SR and VTT in general. People have already mentioned a couple of examples in the comments above. OS/2 Warp's SR feature is another; it was also widely advertised in the US.
So no, saying "hey, some SF writers described something vaguely like this" is not a plausible prior-art challenge to a patent. Nor is some technology that accomplishes something broadly similar. The question is how it's done; what's patented is the method, not the outcome.
Nor is it necessarily obvious to the practitioner. If you do all your SR with Hidden Markov Models, then Support Vector Machines are hardly obvious to you, are they? (No, they are not.)