[6234] in cryptography@c2.net mail archive
multi-party untrained unconstrained speech transcription
daemon@ATHENA.MIT.EDU (Julian Assange)
Fri Dec 10 09:43:10 1999
To: cypherpunks@algebra.com
Cc: proff@iq.org
Cc: cryptography@c2.net
From: Julian Assange <proff@iq.org>
Date: 11 Dec 1999 01:37:52 +1100
Message-ID: <wxg0xavp33.fsf@suburbia.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
http://speechbot.research.compaq.com/
The "transcript" that is output by the speech recognition software
(and shown in small extracts on the Results and Details pages) rarely
matches what was spoken exactly, and often often does not read very
well. Because different people speak at different rates and with
different degrees of clarity, speech recognition software does not
correctly interpret every word. However, research has shown that
meaningful words are recognized with a high degree of accuracy, and
that even when a word is missed, it will most likely be recognized
when it is spoken somewhere else in the program.
