Voice Monitoring System for Airline Pilots.
Chaos theory tells pilots when to take a break.
AIRLINE pilots and air traffic controllers who are too tired to
work safely could soon be identified automatically, thanks to
a Japanese monitoring system that analyses voice patterns
for signs of fatigue.
"Our system is able to detect tiredness in test subjects
10 to 20 minutes before the subjects themselves notice it,"
says Kakuichi Shiomi, chief researcher at the Electronic
Navigation Research Institute near Tokyo, which developed
the system.
Human error currently accounts for around 80 per cent of all air accidents
worldwide. "Crew fatigue is a very real problem, especially on long-haul
flights crossing many time zones," says Shiomi.
Called a fatigue and drowsiness predictor, the system uses the mathematics
of chaos theory to compare changes in the voices of wide-awake, alert
people with those of fatigued people. The change is known to be related
to a drop in blood pressure when people are tired-but it is very subtle.
So Shiomi's team at ENRI, a division of the Japanese Ministry of Transport,
had to come up with a way of analysing speech that brings these small
changes into sharp relief.
The system takes advantage of the fact that many biological signals,
including heartbeat and blood pressure-have a fractal structure.
Like a coastline or a lightning flash, the pattern of the signal is equally
jagged no matter how closely you zoom in on it. But with voice signals
there's one important difference: the fractal structure changes when
the speaker is fatigued.
The changes are hard to spot by simply watching the voice waveform,
so the ENRI team wrote software that converts a voice signal's fractal
structure into a graphical pattern that magnifies subtle, small changes.
Alert voice pattern has a well-ordered trace.
Drowsy voice pattern is chaotic and spiky.
To test their system, the researchers processed recordings of subjects
as they concentrated on reading or doing maths. When wide awake
and alert, their voice fractal graphics are well defined (see Diagram).
But after 20 to 30 minutes of reading, they become noticeably more
jagged. In a cockpit or air traffic control centre, an image processor
could recognise the fatigue pattern and raise the alarm when staff are
becoming fatigued, signalling that it's time for a more alert colleague
to take over.
Tests of the fatigue predictor are now planned on real pilots and air traffic
controllers. If these are successful, Shiomi envisages using the system
on other transport systems, such as oil tankers, where the verbal
commands of drivers and pilots can be monitored.
_______________________________________________________
From an article in the New Scientist
by Peter Hadfield and Paul Marks
Web: www.newscientist.com
December 2000
