The February 2024 issue of IEEE Spectrum is here!

Close bar

Coughing Up Who You Are

West Virginia researchers believe the human cough may be a unique biometric identifier

3 min read

17 November 2004--Fingerprints are currently the only reliable and easy way to tell identical twins apart. Now, researchers have found something else that could work--coughing. In an initial investigation, Jeremy Day, an engineer at the National Institute for Occupational Safety and Health (NIOSH), succeeded in training software to tell people apart by their coughs. The network passed the twin test by correctly identifying Jeremy and his twin Joshua, who has a slightly different sounding cough. "I tend to wheeze a little," Jeremy Day says.

Day and his colleagues in the Developmental Engineering Research Team at NIOSH's Morgantown, W.Va, laboratories recorded voluntary coughs of people in their research group. They then chose various sound signal and airflow parameters such as length, power, volume, acceleration, and variance to train the neural network. They presented their initial findings on 14 October at the Biomedical Engineering Society's 2004 Meeting in Philadelphia.

The researchers hope cough identification can eventually become a biometric technique, such as fingerprint and iris scanning, whereby an input pattern is compared with an existing database. Day also envisions a more practical application--combining cough identification with breath analysis for medical examination purposes or drunk-driving tests. The project is in a juvenile stage, though, and it will be a while before someone can get through a security check by simply coughing into a microphone.

Biometrics provide an automated method of recognizing a person based on either physiological characteristics, such as the iris or face, or behavioral characteristics, such as handwriting. While fingerprinting is widely used in law enforcement and financial organizations, biometrics in general haven't found a wide circulation yet, says Maud Meister, a consultant at the New York City�based International Biometric Group, which provides business and technical consultation to the biometrics industry [see "Biometrics Boom"]. While these technologies are still emerging, she adds, "it's always important to push the forefront of what is possible."

Since he started working at the NIOSH in 2002, Day has been investigating the possibility of using cough signals to detect obstructive lung diseases such as asthma and chronic bronchitis. Current techniques involve measuring obstructions in people's air pathways by making them inhale to their full lung capacity and then exhale forcefully. "This can be really stressful for older people with the disease," says Day, "whereas everyone can cough easily."

So Day and his group set up their recording equipment in a pulmonary clinic and looked for volunteers. Each volunteer was asked to cough into a tube three times. The tube was attached to a pneumotachograph, which measures airflow to and from the lungs, and a microphone, which translates sound pressure waves into electric signals. Upon analysis of these measurements, the researchers found that it was possible to distinguish people with and without a disease; the engineers' analysis always matched the doctor's prognosis.

While looking at the various recorded patterns, Day noticed that a person's flow and pressure graphs always looked the same every time the person coughed. This made him wonder if every individual's cough was reproducible, and if so, whether it could be used for identification.

In fact, the recorded signals of 14 people in the NIOSH engineering team who volunteered as subjects turned out to be reproducible. Day recorded each person's cough between 6 and 110 times, and used half of these to create training data. The other half was used for testing. For the training data, he chose a total of 60 sound-pressure-wave and airflow parameters and used a data reduction method called principal component analysis to reduce these to the 20 most significant components. These were used to train a software-based neural network. "Twenty parameters is still a lot," Day admits. "Ideally the fewer the parameters, the better, but we found that 20 gave us the most accurate results."

Out of 181 coughs tested, the network identified 179 correctly. The results look promising, but Day says there is a long way to go before his idea can become a practical application. For one thing, the engineers need a lot more data to be sure the technique works. The system also needs to be much smaller. Right now it requires two separate computers along with the pneumotachograph and microphone. "We need a smaller system to gather the information," says Day. "We've already thought of ways to do it."

Most important, however, there are inherent problems involved with cough as a biometric identifier. "If someone gets sick, the system will have some trouble identifying them," Day points out. "The volume changes, and fluid in the lungs causes sound changes."

To IBG's Meister, the limitations of cough identification seems very similar to those of voice recognition, another behavioral biometric. "Voice systems have been scrutinized for the fear that false voice matching could go undetected," she says. Besides, if a person is ill or has a hoarse voice one day, the system's accuracy would be affected and it could lock out the individual. Meister also expressed doubt that users would be comfortable having to cough to identify themselves.

This article is for IEEE members only. Join IEEE to access our full archive.

Join the world’s largest professional organization devoted to engineering and applied sciences and get access to all of Spectrum’s articles, podcasts, and special reports. Learn more →

If you're already an IEEE member, please sign in to continue reading.

Membership includes:

  • Get unlimited access to IEEE Spectrum content
  • Follow your favorite topics to create a personalized feed of IEEE Spectrum content
  • Save Spectrum articles to read later
  • Network with other technology professionals
  • Establish a professional profile
  • Create a group to share and collaborate on projects
  • Discover IEEE events and activities
  • Join and participate in discussions