speech recognition

[wviperw]

So, does anybody know how close we are to mainstream speech recognition(SR)? As I understand it, it will take massive amounts of computational power.

What exactly is the computational power needed for though? I am guessing that it is needed to rapidly search through the entire lexicon for every word right? Also updating the polling interval would improve the accuracy. But how much processing power is needed for the functions such as the natural language engine? What improvements to computing will help the most, 64bit arch., wider bus, what?

 

[Killbat]

I think we could do natural speech recognition now, but it would have to be in hardware, based on DSPs and custom logic circuitry. In other words, it would not be commercially viable, but still possible.

 

[wviperw]

commercially viable is the key. Once it is available to the average small business it should be a whole new ballgame. And once it is available to the average household it could change a lot.

 

[Shalmanese]

with a bit of clever programming, you could bring down the system requirments quite a bit by working out common word combinations and plunking in a neural net which can learn your particular nuances in speech.

 

[konichiwa]

Some of the very high end software programs (the "enterprise" (or something like that) versions of Dragon NaturallySpeaking) are VERY impressive.

 

[TravisBickle]

sorry Konichwa, I must have misread you. did you mean to write "very CRAP"?
maybe in 2 years!