> Does anyone know if any of this is robust enough to convert taped speech to
> text - i.e. taped lectures?
The claims for "continuous" speech input still need generous
interpretations. Gaps, even though brief, are needed if many phrases
are to be successfully unbundled into words. Voices still have to be trained
(ie learned by the software), which is not going to happen with a
lecturer, unless s/he is feeling very supportive. The input
conditions have to optimal as even variation is volume level can
affect the error rate, thus the typical portable recorder is hardly
likely to provide a signal that would be acceptable even given
thorough voice training by the source voice, and the need to walk
around, face different ways, etc. would render the signal well nigh
impossible. I suppose that theoretically, a voice which had been
learned, recorded under excellent studio conditions, speaking to a
pattern known to be acceptable to the software, then reproduced to a
hi-fi standard aught to achieve a success rate comparable with the
same thing done live, but this hardly equates to the typical lecture.
I had hoped that the more restrictive condition, ie the student
making notes to themselves using a portable tape-recorder during the
day, with the tapes played back through a decent hi-fi system at home
might possibly work, but our systems manager tells me that evidence
so far makes even this still unlikely.
Dave Laycock
Head of CCPD, Chair of NFAC
Computer Centre for People with Disabilities
University of Westminster
72 Great Portland Street
London W1N 5AL
tel. 0171-911-5161
fax. 0171-911-5162
WWW home page: http://www.wmin.ac.uk/ccpd/
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|