|
Features: March 4, 2002
Teaching Computers To Replace Lost Sounds
Mike Savic, professor of electrical, computer,
and systems engineering, is teaching computers to reconstruct
damaged speech and recover missing sounds, some of which
cannot even be heard.
No, he won't be able to fill in the 18 missing
minutes of the Watergate tapes.
Mike Savic, professor of electrical,
computer, and systems engineering, is teaching computers
to reconstruct damaged speech and recover missing sounds,
some of which cannot even be heard.
|
|
What his research will do is improve verbal
communications for the military, improve hearing aids, and
possibly find lost notes and words in music and handwritten
text.
This project stems from a previous project
that allowed computers to identify languages through distinguishing
the patterns of speech and sound of the language. Savic
took that knowledge and reversed it—knowing a language
lets him program the computer to identify typical sounds
of the language and the transitions between them. He then
uses probability to determine the sequences of missing sounds,
which reconstructs the damaged speech.
Applications for this technology abound.
For example, when a pilot communicates with the control
tower, a word or two may be drowned out by thunder or some
other loud noise. Almost instantly, the computer would reconstruct
the words to ensure safe communication between the airport
and the airliner. Hearing aids could connect to a chip and
work in the same fashion.
Savic envisions that the technology would
be useful to historians as well and would not be relegated
to sound analysis alone. If historians had a page of an
old handwritten text with bits missing, and they knew the
author, the computer could reconstruct the text in the same
way that it would reconstruct speech.
Savic is working with graduate student Mike
Moore and several undergraduates on the project, which is
funded by the U.S Air Force. A patent was recently filed
for this technology.
|
