Unfortunately speech recognition systems need quite a meaty machine to run successfully. If you are serious about using the software then the first thing to get right is your hardware setup. Failure to do so will just cause lots of unnecessary frustration which may cause you to abandon the software altogether before you realize its full potential.
There are 4 bits of hardware that you should not compromise on; they are processor speed, memory, the soundcard and the microphone. A typical system for a serious user would look something like this:
Pentium III 550 [*1] or faster, 384 MB RAM, a good quality soundcard (I use the Turtle Beach Fiji) and a good solid microphone (I use a ANC 700).
There are many good quality soundcard and microphones available for use these days. The ones listed above are the ones that work for me.
Apart from its excellent input quality, the ANC 700 microphone also has a number of features that I find extremely useful. First, I've broken many mics by dropping them or sitting on them, the ANC 700 is a very sturdy mic, which stops it breaking easily! It also has a solid yet flexible boom which keeps the mic in a constant position; it has a mute button and can be detached from the machine without fully unplugging it from the PC. This microphone also has an Active Noise Canceling element; this lets you listen to music in the background without the background noise affecting your recognition too much!
The ANC-600 is basically the same as the ANC-700 in that it uses the same mike element - the only difference is that the ANC-700 looks a hell of a lot more sexy!
[*] These recommendations are for version 4.0 of NaturallySpeaking.
[*1] or similar speed Athalon.
Darran Bryant dnb1002@cus.cam.ac.uk