29 Oct
1998
29 Oct
'98
8:54 a.m.
On Thu, 29 Oct 1998, Fred Cisin (XenoSoft) wrote:
Even the task of parsing the input stream to separate
the words from each
other has remained an elusive goal. That is a loarge part of why the
current "dictation" systems remain unsuitable for transcription, closed
captioning, aids for the deaf, etc. Even now, the best systems available
require intense interaction between the speaker and the program.
Yup, it's a *very* difficult problem. Basically, there's not a
straight-forward mapping between sound waves and what we hear. Even once
you do all of the relatively simple signal processing and word
recognition, you're left with the basically impossible task of context
analysis (which requires almost all of our brain to get right, and we
still blow it about 1% of the time).
-- Doug