Voice AI beats RAT as an instant covid test

Vocal changes during infection are a promising new avenue for fast diagnosis.

---

A cheap, reliable and fast way to tell if you have the spicy virus?

That was the rapid antigen/lateral flow test dream – a dream that soured for many of us who, once the tests were finally available, were betrayed by a false negative. (Or three false negatives, in The Back Page’s case. What a complete waste of good nose cells.)

So the idea of a much cheaper, even easier and – most importantly – far more reliable way to test for covid sounds too good to be true.

But that’s the implication of work presented at the European Respiratory Society International Congress, held this year enviably in Barcelona, and preprinted at arXiv.org.

A team led by Wafaa Aljbawi from Maastricht University trained an AI to detect covid from a person’s voice, using a large set of voice samples from this Cambridge University crowd-sourced project. Each voice sample donor in the dataset had to cough three times, breathe deeply through their mouth three to five times, and read a short sentence three times.

Ms Aljbawi’s team used something called Mel-spectrogram analysis to break down which vocal variables are the key to spotting covid infection, she told the congress.

The researchers built a range of models and tested them all, and the one that performed best achieved 89% sensitivity and 83% specificity.

Compare that to the average RAT sensitivity cited by Ms Aljbawi of just 56%, and it’s a frankly staggering improvement.

Add the fact it costs basically nothing, can turn around a result in under a minute and involves no swabbing of sensitive bits, and they could be on a real winner.

The successful model, the oddly named Long-Short Term Memory, was based on neural networks, mimicking a brain’s way of finding relationships in data. The study, as already noted, is only at preprint stage and the team is doing more work to validate the model with larger numbers.

If it checks out, such a test could be deployed, as Ms Aljbawi said, “at the entry points for large gatherings, enabling rapid screening of the population”.

The only unfortunate thing could be that by the time this tech can be mobilised, people will have ceased to care enough about covid to wait even a minute for a result that might spoil their day and week.

If you see something that makes you testy, tell penny@medicalrepublic.com.au.