Yes, COVID-19 has made the year 2020 a living nightmare and yes, it’s still here… we think – Is it? No? Yes? Well, at least the flu is no more. You already know how some people are asymptomatic (meaning HEALTHY), which is used as an additional method to keep you in fear, but I digress. Now, your healthy cough can (allegedly) reveal if you’re haulin’ virus bits around unbeknownst or not.
Magic A.I. 8-Ball Says… You Diseased
MIT researchers Jordi Laguarta, Ferran Hueto, and Brian Subirana have developed an artificial intelligence model that is designed to distinguish COVID-19 asymptomatic individuals from audio recordings. By letting the A.I. analyze thousands of different coughs and speech patterns, the team was able to calibrate it to recognize 98.5% of asymptomatics who showed no signs of the virus but tested positive for it. Could they have been false positives? Who really knows.
It builds on the work multiple research groups were conducting even before the pandemic hit; only instead of testing for COVID-19, people were looking to diagnose patients with asthma, pneumonia, and even Alzheimer’s (weakened vocal cords can be a symptom of the aging disease).
According to the official MIT news site, the team combined a total three neural networks:
The first neural network is a trained general machine-learning algorithm called ResNet50, which is programmed to distinguish speech patterns. Sounds like “mmm” or “err” can tell how strong or weak a person’s vocal cords are, and these could show the first signs a person is struggling to speak due to being infected. Or is it just phlegm stuck in your throat or that you had too many Saki bombs? Inconsequential.
The second neural network focuses on how people convey their speech. Emotions play a big role in how we talk, and this network tries to classify how you’re feeling when you’re talking. Frequent pangs of frustration and even no emotion while speaking could be telltale signs that a person is suffering from COVID-19, and this second network was made to sniff these out. Are there other possible feelings here? Psshh – please.
Lastly, the third neural network is trained on a wide database of coughs. Big coughs, small coughs, strained coughs, and suppressed coughs – all these and more are classified under this network where it then identifies if you have any discernible changes in your respiratory system. Ahhh, the life and times of a cough.
These three neural systems, combined with an algorithm that detects muscular degradation, make up the final A.I. built specifically for detecting asymptomatic people who may be carrying COVID-19. It can’t diagnose those who already have symptoms of the disease, but it can distinguish an asymptomatic cough from a normal healthy cough, so we should accept that it’s completely accurate.
The next step for the MIT team now is to incorporate their A.I. into an easy-to-use app which can be shared on a large enough scale. The plan is to provide free screenings for people who may be suspected of carrying COVID-19 but don’t necessarily show any physical symptoms. All you have to do is download the app onto your phone, cough into the speaker, and confirm whether or not you may be asymptomatic.
Who’s ready to turn their health over to A.I. and an app? Raise of hands? Anyone? Anyone? “Cough into ze app or it is curfew for all of you!!”
You can find Laguarta, Hueto, and Subirana’s paper on their cough and speech analyzing A.I. on the Institute of Electrical and Electronics Engineers (IEEE) Engineering in Medicine and Biology Society (EMBS) webpage.
