Thank you both very much.  As usual DonationCoder.com comes through with flying colors.
NR

Well, what do you know, something I know about.

Here's how I do transcriptions from various sources. First off, Dragon is the best program for this, rant or rave, hands down, I've used them all since L&H, Via Voice etc. So unless you have a way to unlock the TV COP SHOW Prop computer that the pretty girl types into and quickly finds the secrets of the universe and the perps/unsubs pet's middle initial, you are going to have a learning curve. So I'll skip right to the short cut.

First, for a recorded file you have never performed any "training for" (I'll assume you've seen the tv commercial where the professors lecture appears miraculously quick and correct as a text document on a computer from 30 feet away (10 meters) using only the microphone installed in the computer cover?  If you haven't, no worries.  It's the same computer as the TV COP SHOW Prop computer that the pretty girl types into and quickly finds the secrets of the universe...  OK.  not going to happen.)

Now Curt (famously) is the only poster who got this right and was kind of "glossed over" and his suggestion about using Audition seemed to end.

Back to the shortcut. You need to process the file before you drag it into 'dragonpad'. The quick way of putting it would be to say, the best file should sound something like Donald Duck.  No, seriously.  Think about it. The less mid range/low and harmonics there are the better the speech engine will do.  I am NOT going to get into the audio analysis of all this here.

Second, my experience is you are better off starting with a .wav file.  Basically cut the lows and mids so it sounds somewhat "tinny". Pull out the pops and hiss and other noise. Slow it to about 67%. Then drag that processed sound file into Dragon or whatever.  (You may need to convert to mp3.)

Here's some summary points:

The clearer, less noisy, better articulated speaking on the original, the more accurate the text file will be.

Having a high quality microphone positioned properly is huge. (see above)

You can get around 80% accuracy. Remember most speech recognition programs spell the words correctly 100% of the time.  They just spell the wrong words.

About this 80%. It's not a distributed %, You may get a complete paragraph with 98% accuracy, and then a sentence completely unintelligible.

If you can get the speaker to dictate a few paragraphs from Dragons training files placing the mic at the pick-up position will greatly help accuracy.

Some recorded situations just don't work.

Hope this helps.  Cheers.

NR
Washington DC

Well, I suppose I'll be chastised (flamed is it?) for this, yet my misspent youth and the anarchistic force in me insists I speak up. I'll make it brief: It's time to burn and loot.

the APP store pirates of the world have done a magnificent job of changing the thinking of a generation, and at the same time performed their public relations legerdemain of hiding the truth. If those who believe the internet, the WWW, the frequency spectrum, DOES NOT belong to corporations that simply declare their size dictates ownership, do not begin now to make a noise that reverberates for decades, kiss donation coding good bye.  Hasta la vista freeware.

Called for above is my philosophy of go big or go away...don't JUST burn...don't JUST loot. However with experience and age and a few wounded causes comes respect for reality. You still need to keep up a fight with many many short jabs and an occasional knockout punch.  Keep in mind the other side is trying the same thing.

You still need to go BIG, but from the inside. And that's my thesis.  I'll leave the specifics to you as to how to go about keeping the next generation of coders from selling out.