Voiceprints May Help Identify bin Laden

Nov. 13, 2002 -- Like a fingerprint, every voice is unique and characteristic of the instrument that creates it.

In analyzing the latest audiotaped message purported to be from Osama bin Laden, investigators can use voice verification systems that compare phrases and words to matching ones in previous recordings.

The programs search for obvious and subtle trademarks of a person's vocal instrument as well as trace commonly used words, intonations and patterns of speech. Although experts say voice identification can never be as exact as DNA or fingerprints or even iris identification, it's a technology that has become more accurate in recent years.

"Your voice is nothing more than signals," explains Jordan Byk of VeriVoice, a voice verification company in Princeton, N.J. "A machine breaks down these signals and measures your individual physiology and how sound passes through it."

Parts of Speech

To identify a person's so-called voiceprint, a computer breaks down the sounds of speech into tiny components. Voiced sounds, unvoiced sounds and plosive sounds make up the basic components of speech sounds.

Voiced sounds — like "aah" — are produced by the vibration of our vocal cords when air is exhaled from the lungs. This vibration produces a wave audio output that is passed through and manipulated by the vocal tract and then radiated at the lips.

Unvoiced sounds — like the letter "f" — are made mostly by the hissing of a steady air stream through the mouth. The vocal cords don't vibrate during these sounds and they can be distinguished by factors like air pressure and tone.

Plosive sounds, like the letter "p" in stupid, are made by completely closing the vocal tract and building up pressure behind the tract and then abruptly releasing it.

Because every person's mouth, larynx and nasal cavities are unique, each utterance they make is also unique. Also, the way people use their muscles when creating sound is imprinted in the voice and can be identified by a computer.

Comparing Voiceprints

Analysts studying bin Laden's voiceprint will be tracing computerized "wave forms" that represent the frequency, cadence, tone, pitch and volume of the voice. These traits are plotted onto a color-coded topographical map.

To identify his voice, they will likely use a topographical profile of his voice from previous recordings. The program will then break down the voice components in the new recording, compare it with the model and calculate the probability that it's a match.

Although voice verification technology has accelerated in recent years, some argue it still a technology with significant weaknesses.

"It's can't discriminate as well as other kinds of verification like an iris scan or fingerprint," says James Glass, a spoken language expert at the Massachusetts Institute of Technology. "Every time you talk, your voice changes depending on the environment, your mood or if you're sick or stressed."

Others point out that variables like the way a tape was recorded and the quality of the tape can skew results and that people can often be better at recognizing recorded voices.

ABCNEWS correspondent John Miller, one of the few American journalists to ever interview bin Laden, said the voice on the new tape "certainly sounds like bin Laden, not a crudely edited tape." Yossef Bodansky, author of The Man Who Declared War on America, a book about bin Laden, said there's no doubt in his mind the tape is authentic.

"It is his language. It is his dialect. It's his voice," Bodansky said.

No matter how certain people may be, Byk argues, final verification should rest on technology.

"When a person hears a voice, they automatically associate it with a familiar voice and then reinforce that memory," he said. "Computers don't have that issue. It's almost impossible to fool a computer."