Coughers identified in algorithm study

It is a development that begs the question, what bodily function will not be turned into a biometric identifier.

Researchers from the University of Washington and Google have written code that reportedly is pretty good at identifying who among a group of subjects is coughing. It could be used in diagnosing and treating cystic fibrosis, tuberculosis and asthma.

It might even work on populations that have a future form of Covid.

Their paper claims that their model correctly classifies the noise 82.15 percent of the time. The researchers went beyond that, using an algorithm trained on speech, something that already has been done. Speech code “performed reasonably well on forced coughs,” according to the paper’s authors.

They instead used multi-task learning. The second task was biometric speaker verification among four subjects from a dataset of natural, in-the-wild coughing.

The algorithm bested humans doing the same task by 9.8 percent, according to the authors.

Ultimately, the researchers found that they can on average outperform human listeners on the four-way classification task with 10 enrollment samples.

A researcher behind a similar paper published earlier this year told Wired that while the algorithms can have practical value for health applications, but may need further development to be used as a biometric.

Article Topics

algorithms  |  biometric identifiers  |  biometrics research  |  coughing  |  Google