On speech recognition
One of the problems of using speech recognition software is the frequent need to make corrections. While the number of corrections needed has significantly declined with NaturallySpeaking version 7, there are inadvertent grammatical errors and various other minor problems that occur with speech recognition, that make one's output somewhat contaminated. In addition, the very act of starting the software, putting on the headset, etc can take time, which means that there is a strong temptation to directly type a few words, particularly in responding to e-mail.
For extremely fast typists like me, speech recognition does act as a barrier. I must therefore force myself to use it, given its beneficial impact on my condition. But very often, my output is a mix of hand typed material and dictated text. So there would be spelling mistakes caused by hand typing as well as grammatical mistakes caused by dictation. Not a very pleasant mix. It can also confuse the reader at times. I find particularly problematic in work at office, where people point out absurd errors that arise from this mix of methods.
PS. I now use Version 9 (sanjeev, may 2007)