Sberbank has developed an algorithm that determines the likelihood of COVID-19 disease by voice, coughing and breathing. This was reported by the press service of the company.
“Sber’s artificial intelligence laboratory has developed an algorithm that within 60 seconds can determine the likelihood of contracting coronavirus infection COVID-19. The presence of the virus is determined based on the results of a short symptom survey and three sound patterns - voice, breathing and cough. Sound files are converted into a spectrogram showing the energy of sound at different frequencies, and then analyzed using a deep convolutional neural network,”the message says.
As specified, only open data was used to train the neural network - these are more than a thousand samples of breathing and coughing sounds collected from diagnosed patients in Russian clinics.
“At the beginning of November last year, we announced our readiness to create such an algorithm and since then have not wasted time in vain. Of course, our model does not yet achieve the accuracy of biological PCR, which is not surprising, but now it has comparable characteristics. At the same time, it makes it possible to make customizable sensitivity, much easier to use, more convenient and, importantly, cheaper. This is not a medical diagnostic tool, but rather a personal daily checker - taking the test and getting the result takes only 60 seconds. In the near future, we plan to create a special application that will become available in the App Store and Google Play. This will make it possible to better tune the accuracy of the model, "said Alexander Vedyakhin, First Deputy Chairman of the Board of Sberbank, who is quoted in the message.
The company expects further improvement in the quality of the model with an increase in the amount of data, including those collected using a mobile application.