pi********@hotm ail.com wrote:
Dragon works with pretty much anything in Windows. What are you
trying
Please excuse the very long sentences. Plus, this post doesn't even
answer the OP's question. Here is my perspective on why speech
recognition technology is where it's at:
The two men who started Dragon Systems used to work at a company called
Verbex. Verbex specialized in speech recognition and their slogan was
"No sooner said than done." A friend of mine had them as supervisors
at Verbex. He even mentioned that one had a statue of a red dragon at
his home in Newton, MA. My friend had done AI research at MIT and had
also worked at Lincoln Labs and BBN. The neural net program he was
paid to develop at Verbex for speech recognition was nowhere near his
later efforts, but it was good enough for those two to leave Verbex
with the code and start Dragon Systems. Around the same time my friend
left Verbex. Working at home, he was able to create a so-called
evolutionary neural net that used speech data alone to create over
fifty layers of neural net. At the fifty layer level he was able to
print out things like vowel patterns. His new results allowed him to
formulate a workable knowledge representation. His new speech
recognition system idea would need no training once constructed. It
would be able to handle several transformation invariants such as
different pitches, local dialects, elision, etc. He even devised
methods for maintaining the highest recognition level possible when
pruning the network for placement on chips. Even problems like
instantaneous translation from one language to another became feasible.
Then he decided not to market any of it.
His next project was a symbolic debugger. I watched him turn pure
assembly code into interpreted c containing symbol names and including
the names of all library calls in a few hours. The debugger would
systematically discover new information and make appropriate
replacements everywhere. He naively submitted a beta version to
Microsoft for evaluation. Instead of hearing back from them he
discovered that they were selling his beta version and didn't even
bother to change the icon. This experience probably solidified his
decision not to market his speech recognition software. The easier it
is for users the harder it is for programmers. Microsoft lost a great
opportunity for making things easy for users. My friend's explanation
many years later of how thoughts work while pointing to the diagram of
his knowledge representation along with his explaination of how
thoughts activate certain node areas in the brain remains the most
incredible thing I've ever encountered. Around 1990 I submitted a
short paper to IEEE and gave a lecture explaining what happened and
giving him credit for his remarkable accomplishment.
James A. Fortune
