VoiceMed Detects The Sound Of The Virus

An international initiative born during a hackathon has developed into an app that can screen Covid-19 symptoms via a simple phone call and report on the risks of infection in real time. The team expects to launch its startup in Luxembourg soon.

Voice recognition

Once connected, callers are requested to cough three times, breathe deeply several times and say a specific sentence in English. Test results are shared with the callers immediately.

“The application can tell how high, medium or low the infection risk is”, explains Cédric Tamavond, one of the three VoiceMed’s co-founders. “Results are sent to callers via SMS. And when affected they can decide to contact their doctor and apply for a Covid-test.

The solution can unmask the underlying pathophysiology (disordered physiological processes associated with disease or injury) based on voice markers that are specific to diseases through voice recognition technology.

“We know that when infected patients are requested to cough, they usually cough more than three times and have difficulties breathing. And when asked to deeply breathe, they will not be able to do it completely and will face breath shortness”, Tavamond diagnoses.

The extra sentence to be pronounced will give further information on patients and more accurate results on their diseases, such as their estimated age (the older they are, the more likely they are to be infected), and geo-tracking (as viruses can mutate and take different forms depending on their localization).

“As the app can support millions of calls simultaneously, we could pre-screen Luxembourg within a few hours instead of the current six months.”

Cédric Tamavond, VoiceMed co-founder

Artificial Intelligence and voice database

To identify the disease, the application uses Artificial Intelligence algorithms and relies on a voice database. For that, VoiceMed has developed a Web-application that collects voice recordings; Around 1,500 voices of Covid-positive patients have been already collected, through partnerships with hospitals and research centers in Lithuania, Spain, Brazil and Nigeria.

A voice model has been created, based on the collected samples; It will serve as a vocal reference to determine whether and how high the risk is. The model, however, still needs thousands of voice samples to gain precision.

“As the app can support millions of calls simultaneously, we could pre-screen Luxembourg within a few hours instead of the current six months”, Cédric Tamavond says. “This could help health and medical facilities to focus primarily on people who are really infected”.

Each VoiceMed’s test will cost around 2 euro: “Compared to individual PCR-tests that costs between 50 and 100 euro, the country could save millions of euros”, Tamavond adds. “Therefore, it would make sense to work with governments and develop the solution together”.

In the future, the app is due to include a dashboard and functionalities that will help governments to estimate in real time the number of tested and infected persons, and then take strategic decisions and actions.

Startup in the making

“Our solution will respect the General Data Protection Regulation (GDPR)”, Tavamond insists. “After the test, we will not keep the information on callers”.

VoiceMed will soon register as a startup with its headquarters in Luxembourg. Tavamond does not know yet, in which incubator the structure will be hosted.

In order to launch a precise product within three months, the future company needs a funding of 50,000 Euros, 15 more engineers and partnerships with Luxembourg-based hospitals and research centers, which could provide more voices.

Launched half a year ago during the online Hack for crisis Italy, the VoiceMed initiative includes around 20 people – machine learning specialists (on cough and speech), medical researchers, Product/Tech as well as Data acquisition specialists – based in ten countries.

Total
0
Shares
Leave a Reply

Your email address will not be published. Required fields are marked *

Related Posts
Total
0
Share