Talking Computers Nearing Reality
From: CNet - July 9, 2003
By: Michael Kanellos

Upcoming products from IBM and Microsoft aim to bring the dream of
conversational computers closer to reality and move the technology out of
niche markets. Microsoft will issue on July 9 the first public beta of its
Speech Server software, a back-end offering that seeks to lower the cost of
deploying automated phone response systems. The software has three components
- a speech-to-text engine to make oral commands comprehensible to computers,
a prompt engine that helps guide callers with pre-recorded responses, and a
text-to-speech engine to translate unfamiliar computer responses or inquiries
into oral communications. Meanwhile, IBM announced plans to bring computers
that can support conversations between two people speaking different
languages out of the laboratory by the end of 2003, and the company's Super
Human Speech Recognition Project aims to make commercially feasible,
exceptionally accurate speech-to-text transcription systems a reality by the
end of the decade. Researchers have redirected their efforts away from
creating machines that can actually converse to systems that can understand
speech as a probabilistic function. For example, Microsoft's Alex Acero
explains that the Yoda speech-to-text engine currently in development
converts voice into written text by analyzing user habits and picking up on
specific sound patterns. Chalapathy Neti of IBM expects many new systems to
be equipped with cameras that can help computers better understand what
people are saying despite heavy background noise by studying lip movements
and facial cues. Companies see much greater potential in embedding speech
technology into cell phones and other devices rather than PCs, a strategy
that speech proponents claim is better aligned to user needs and serves
growth markets. 

http://news.com.com/2100-1008_3-1023966.html