New York Times

Experts Say a bin Laden Impostor Could Fool a Lot of People

By Michael Moss 
November 14, 2002 

The government's assessment so far that it cannot be absolutely certain that the audiotape broadcast on Tuesday was recorded by Osama bin Laden does not surprise experts in the field of voice authentication.

The science of using computers and linguists to identify individuals by their speech has improved dramatically in the last several years, but still involves considerable guess work and speculation, the experts say.

"If an impostor wanted to put together a tape, he could fool a lot of people," said Michael J. Thieme, director for special projects with the International Biometric Group, a New York consulting firm.

Experts said that the most successful impostor would probably be a member of Mr. bin Laden's immediate family, though mimicry is often detected by the new technology used in voice authentication.

In analyzing the latest message attributed to the leader of the Qaeda terrorist network, analysts with the Central Intelligence Agency and the National Security Agency were hampered by sound quality, senior Bush administration officials say.

The message was recorded or re-recorded over a telephone line, but American officials said they were not certain whether Al Jazeera had recorded it from a telephone line or whether the telephone connection had been made with an intermediary before the tape was handed over to the Arab network.

Playing it over a telephone line would have wiped out the high and low frequencies that help distinguish voices. But even if the quality had been superb, the government analysts would have faced pitfalls in trying to verify the voice.

The vocal tract physically differs from one person to another, but unlike verifying someone's identity through fingerprints, voice recognition is indirect.

Government analysts would have begun by compiling a set of recordings that Mr. bin Laden is known to have made, called a library in the parlance of voice authentication. Most of these recordings came from videotapes in which it is clear that Mr. bin Laden was the speaker. This earlier material provided a wide range of his speech and varied background noise with which to compare the new tape.

With the previous recordings as a template, the government analysts would have then fed the old and new tapes into a computer using software that most probably was designed expressly for the intelligence agencies. The program converts speech into bits and chunks, and then computes the degree to which the two sets of tapes match.

Intelligence officials have said the resulting score means there is a 90 percent probability that the voice on the new tape is indeed Mr. bin Laden. By comparison, fingerprinting done by law enforcement officials is accurate in the order of 99.99 percent or even higher.

But there are no scientific standards in estimating the accuracy of voice authentication, and the raw computer scores would have been subject to interpretation, Mr. Thieme said.

Even while the computer was doing its work, the government would have relied on linguists to assess the tape. They would have been looking to see if it contained phrases and speech patterns known to have been used by Mr. bin Laden. Voice experts would also listen for inflection and tone and other speech characteristics that vary from one person to another.

"This is a very active area of research," said Larry Heck, director of speech research and development at Nuance Communications, a leading voice authentication company. "The types of words people choose, how frequently they use them, how they pronounce certain words, and the melody or pitch. All those things stacked together separate people out.

"It's not only the words he says but how he says them that counts, and some people have better expertise at this than others."

Someone trying to fake the voice of Mr. bin Laden might begin with the voice of one of his sons, said Lonnie Smrkovski, the retired chief of the voiceprint unit of the Michigan State Police. The son's voice could then be edited with commercially available computer software and mixed with background noise to mask the differences with the father's voice.

Various frauds have been attempted over the years, occasionally in an effort to rewrite history. A common technique is to string together words from previous speeches.

"Years ago, I saw on the news that somebody was selling a tape of Richard Nixon in which he was claiming responsibility for Watergate," Mr. Smrkovski said. "I ordered it and took a look and what they had done was alter the original tape, adding in background noise. But you could tell the words were spaced too closely together."

However, the latest message attributed to Mr. bin Laden included words that experts said had not been found in some of his previous speeches, such as "Pharaoh" and references to events as recent as the Bali bombing and the attack on a French oil tanker off Yemen.