Hi Fred,

Thanks for the clarification. 

I found a video interview (Robert Scobble is the interviewer) of John Pollard (president of Jott Networks) and Shreedhar Madhapetti (VP of Engineering, also with Jott).  It is 30 minutes long.  They start talking about the hardware architecture at 11:40 minutes (it's based on .Net ... they are both ex-Microsoft), and at 13:00, they start talking about their recognition technology. 

They say they started looking at Nuance and Microsoft for speech recognition.  But they wanted to create a system that did not require any voice training, with an essentially unlimited vocabulary (they did not want to limit the user to a specific grammar set or syntax).  In addition, they wanted people to be able to use the service from cell phones and/or in noisy environments (car, between meetings, bar, ...).  This presented them with a very challenging speech recognition problem.

Because of this, they decided they needed to collect a lot of speech data up front, so they started off doing just manual transcriptions (leveraging the current overseas transcription services that are already in existence), with plans to increase the amount of automation they do. 

Since some speech recognition engines can give a confidence level of how well this can transcribe a certain segment of speech, this permits them to use a hybrid model of machine-made transcriptions with a fall-back to manual transcription if the confidence levels are too low.  As time goes on, they plan to go to a higher and higher mix of automation to a fully automated state.

Ken