I want to touch on a point that I missed in my previous message.

Natural-language speech recognition programs rely heavily on
word-sequence frequencies to achieve accuracy. For example, if the
program hears you say something that sounds like "I doughnut," it is
likely to conclude that you said "I do not," because that is a much
more likely sequence of words.

When you contemplate using speech recognition to enter authors' names,
search strings, etc., you have not only passed beyond the scope of
limited-vocabulary speech recognition; you have passed beyond the
scope of natural-language speech recognition, too. You are asking the
software to recognize speech that is not only drawn from an unlimited
vocabulary, but also lacks reliable word-frequency clues. This is
beyond the capability of any currently available commercial software,
even running on a full-fledged personal computer with a high-quality
boom microphone, used by an experienced operator under nearly ideal
conditions.

My email address is LLM041103 at earthlink dot net.