November 14, 2002, Thursday
Copyright 2002 Daily News, L.P.
Daily News (New York)
November 14, 2002, Thursday SPORTS FINAL EDITION
SECTION: NEWS; Pg. 8
LENGTH: 389 words
HEADLINE: IS VOICE OSAMA'S? EXPERTS
BYLINE: By DEREK ROSE DAILY NEWS
STAFF WRITER With News Wire Services
Computerized analysis should be able to determine conclusively whether the voice
on the latest Osama Bin Laden tape is really the terrorist mastermind's, audio
experts said yesterday.
"I think they'll be able to say one way or the other, with a scientific degree
of certainty," said Lawrence Kobilinsky, professor of forensic science at John
Jay College of Criminal Justice.
Even a skilled impersonator should be no match for a computer analysis that
could compare vowel sounds on the tape to verified recordings of Bin Laden.
Those sounds are formed by the movement of the lips, tongue and jaw - physical
traits impossible to mimic.
"Someone doing a good Bill Clinton [impersonation] wouldn't confuse these
algorithms at all," said Dan Ellis, an assistant professor at Columbia
University who studies speech recognition.
Rich Sanders, an audio forensic specialist with the University of Colorado, said
he once used computer analysis to compare a recording of former President
Richard Nixon with one of impersonator Rich Little. The differences were
instantly obvious in the computer analysis.
"It was pretty amazing, really," Sanders said.
Because the computers compare only individual words, there's no way to tell
whether the Bin Laden tape was created by splicing together earlier recordings,
audio experts said.
Some analysts aren't so sure the technology is foolproof.
"If you have someone who can read like Bin Laden, they can have him read it,"
said Robert Berkovitz, a speech analyst with Cambridge, Mass.-based Sensimetrics.
"People can change their voices so easily."
Steve Cain, an audio expert who analyzed Gennifer
Flowers' tapes of Clinton for CNN, said a low-quality recording could hamper the
Voiceprint identification is done using a spectrogram, a computer
snapshot of a person's voice.
A spectrogram measures unique qualities, such as pitch, volume,
inflection, dialect, articulation, resonance and breath patterns. All of these
are affected by the physical attributes (nasal and vocal cavities, shape of the
vocal cords, muscles of the mouth) of the speaker.
Investigators identify key phrases on a known recording of a person's voice. The
mystery voice is then compared with the original recording, using as many as 20
examples of those key phrases.