Detecting COVID-19 is not easy, because the symptoms it causes vary greatly from one person to the next. If you have some symptoms such as a sore throat, fever, cough, or loss of smell or taste, you’ll know that you need to get yourself checked.
That’s why companies like Vocalis are developing solutions to pre-screen people using simple technologies such as voice and mobile apps to predict if someone has COVID-19.
What problem is it solving?
Screening folks for coronavirus or any other disease is difficult unless you go to a test center in person. So Vocalis has developed an AI that identifies voice characteristics of patients having various diseases without needing to visit a hospital or stressing their local healthcare systems.
Plus, screening methods such as temperature guns can’t detect asymptotic patients. Vocalis’ solution could help in those situations.
[Read: New AI tool can diagnose COVID-19 just by the sound of your coughs]
Where’s the AI?
Vocalis collects a lot of data related to a person’s voice. It has developed a technology that identifies characteristics of a potential disease based on the analysis of voice collected from a person.
The firm’s algorithm extracts 512 features from a voice sample and makes an image — a spectrogram — with which it can diagnose possible diseases.
When the company aims to help detect a specific disease, it first holds clinical trials to recognize what kind of voice recordings it would need to collect to identify it, and what signatures in the audio it will need to pinpoint.
For instance, to detect COVID-19 the AI needs to identify characteristics of symptoms such as fever, headache, and shortness of breath.
How was it trained?
Vocalis has made sure that the AI is language agnostic. It asks patients to count from 50 to 70 to train the algorithm. Dr. Shady Hasan, the co-founder of Vocalis, said that this is to preserve a patient’s privacy and not to give away anything about their identity.
He added that this also aids them to maintain consistency of samples as compared to a dataset with spoken phrases. In turn, this helps the firm identify vocal biomarkers related to a specific disease easily.
[embedded content]
The company’s pre-training dataset includes over 275,000 different speakers in various languages to make it universally accessible.
Let’s talk about some stats
In February, Vocalis — an Israel -based company founded in 2019 —entered a joint study with the Municipal Corporation of Greater Mumbai in India that tested 2,000 participants.
They spoke multiple languages, including English, Hindi, Marathi, and Gujarati. It installed its VocalisCheck tools on smartphones to gauge the efficiency of algorithms. At the end of this project, researchers found that the algorithm was 81.2% accurate in detecting COVID-19 — a lot of them being asymptomatic patients.
Vocalis’ tools are currently deployed in the US, Chile, Indonesia, South Africa, Romania, and Luxembourg in the public and private sectors.
What’s next?
For the next steps, the firm wants to acquire more datasets to make its COVID-19 detection more robust and accurate. Plus, it’s looking to commercialize the solution and make it available for more regions.
Published April 2, 2021 — 08:18 UTC