13 April 2009 by Paul
Marks
YOU are the victim of identity
theft and the fraudster calls your bank to transfer money into
their own account. But instead of asking them for your personal
details, the bank assistant simply presses a button that causes the
phone to produce a brief series of clicks in the fraudster's ear. A
message immediately alerts the bank that the person is not who they
are claiming to be, and the call is ended.
Such a safeguard could one day be
commonplace, if a new biometric technique designed to identify the
person on the other end of a phone line proves successful. The
concept relies on the fact that the ear not only senses sound but
also makes noises of its own, albeit at a level only detectable by
supersensitive microphones.
If those noises prove unique to
each individual, it could boost the security of call-centre and
telephone-banking transactions and reduce the need for people to
remember numerous identification codes. Stolen cellphones could
also be rendered useless by programming them to disable themselves
if they detect that the user of the phone is not the legitimate
owner.
Called otoacoustic emissions (OAEs),
the ear-generated sounds emanate from within the spiral-shaped
cochlea in the inner ear. They are thought to be produced by the
motion of hair cells within the outer part of the cochlea.
Typically, sounds entering the ear cause these outer hair cells to
vibrate, and these vibrations are converted to electrical signals
which are transmitted along the auditory nerve, allowing the sound
to be sensed. Crucially, these cells also create their own sounds
as they expand and contract.
That's because "hearing is an
active process - the ear actually puts energy into the incoming
sound waves to replace energy lost as sound is absorbed by the
ear's structure", says Stephen Beeby, an engineer at the University
of Southampton, UK, who is leading the research. "This process helps
us hear things we otherwise would not, but as a result some of the
energy added by the hair cells escapes as OAEs."
Predicted in the 1940s but not
detected until ultralow-noise microphones were developed in the
1970s, OAEs can be provoked when a series of clicks is played into
the ear. The returning sound emissions comprise signals of between
0 and 5 kilohertz, and vary in amplitude. Click tests are already
used to check newborn babies' ears for signs of hearing
difficulties, since the OAEs are weaker if the inner ear is
defective.
What sparked the interest of Beeby
and his colleagues is the fact that the power and frequency
distribution in the OAEs provoked by specific series of clicks seem
to be highly distinctive, driven by the internal shape of the
person's ear. "Anecdotally, audiologists say they can tell
different people apart - men, women, even people of different
ethnic origins - by the profile of the widely varying types of
emissions the clicks evoke," he says.
So with funding from the UK's
Engineering and Physical Sciences Research Council, Beeby's team is
attempting to work out if OAE patterns can be used in biometry,
like iris scans or fingerprints. "In the controlled conditions of a
lab, everybody's emissions are indeed different, but whether this
is a practical way of telling people apart as a real-world
biometric still needs a lot of work," he admits.
There are a number of problems
that must be dealt with, he says. In subjects that have been
drinking alcohol, for example, emissions are deadened. And
different drugs alter the amplitude of OAEs, as do ear infections
or wax build-up.
If they succeed by the project's
deadline in mid-2010, they hope to interest electronics firms in
making headsets or cellphones with a supersensitive microphone in
the earpiece. The rest is done with software, says
Beeby.
Establishing a new biometric,
however, is a huge task. Tony Mansfield, head of biometrics
assessment at the UK's National Physical Laboratory
in Teddington, says the team will have to prove not only that their
technique has a low false-match rate, but also that a person's
recorded OAE will match their OAEs over the long term. "It has to
be able to reliably recognise people over long time periods," he
says. "For example, a fingerprint taken from a 20-year-old is still
valid when they are 60."