[kwlug-disc] Speech to text
jvj at golden.net
jvj at golden.net
Wed Feb 23 14:30:25 EST 2011
Back in the mid-1980s a US division of multi-national I worked for had
experimental voice-recognition technology funded by one of the 3 or 4
letter government agencies south the border. The technology was
"speaker-independent" and I helped demo it, as part of a PBX or Telco
switch, at a tradeshow in Singapore. Attendees came from Asia, Japan (is
that not Asia?), India, the Phillipinesm, etc. The technology worked
well with all of the different voices, without training. But the
recognition vocabulary was limited, e.g. "call mister smith" "dial 4 3 2
1". Pacing the users speech into distinct phonemes was required.(Think
syllables with dead air between them.) And is unnatural. I do not know
what happened to this after the company and I parted ways.
Also, in the 1980s, at a Comdex show I saw a TI option board (and
software) that implemented speaker-dependent voice-recognition
technology. The system described by Charles sounds to be similar to the
TI option board (and software).
The problem: The TI option was $10,000 in 1988 US Dollars and required
a top-of the line AT class computer, also circa 1988.
Summary: Two technologies: "speaker-independent" and
"speaker-dependent". One requires user training. The
"speaker-independent" technology required lot of laboratory prep and I
do not know if it ever made it out the lab. And I think that both have
limited recognition vocabularies.
And that is about the extent of my knowledge of these technologies as
my work in the field ended in about 1989.
On Wed, 23 Feb 2011 12:01:33 -0500, Charles M <chaslinux at gmail.com>
> Back in 1996 my brother Michael and I went to Comdex in Toronto and
> sat down with the folks in the IBM OS/2 camp. They were promoting
> 4 which had speech recognition built-in. I believe IBM had bought
> Via-voice and integrated it in.
> Both Michael and I sat down and trained on it for 45 minutes. We also
> were around for other people who trained on it.
> His voice was fantastic, he could easily move all the windows around
> and recognition was really quite good. I was able to move windows
> around, but found recognition was not nearly as good as my brother's.
> Others had pretty much the same experience, their voice just didn't
> register as well as Michael's.
> kwlug-disc mailing list
> kwlug-disc at kwlug.org
More information about the kwlug-disc