The Israeli company, Vocalis, says it can detect a potentially infected person through a voice recording.
Nasopharyngeal tests, soon to be obsolete? In May 2020, researchers from Massachussetts Institute of Technology (MIT) have published a study which showed the possibility of diagnosing the virus, by analyzing the way of breathing and of speaking. This is the bet taken by the Israeli company, Vocalis, founded in 2019. It has developed an application, called VocalisCheck, capable of detecting Covid-19 via a voice recording.
Artificial intelligence at the heart of the application
Covid-19 reaches the respiratory system, especially in the diaphragm, which can cause variations in speech. Aspiration, stability, frequency or speed are all elements that can be analyzed using the voice. In effect, when a user opens the app, they have to record a voice sample counting from 50 to 70 and answer questions about their symptoms. For Dr. Shady Hasan, co-founder of Vocalis, this technique does not reveal the patient’s private life. Once the recording is finished, it is transformed into a spectrogram, that is to say in the form of an image. A total of 512 characteristics are analyzed to find voice signatures that evoke the coronavirus. Artificial intelligence and machine learning technologies then compare the recording with 275,000 extracts from around the world. This operation only takes one to two minutes before displaying the final result.
Reliability over 80%
The first results of the application are conclusive. Last February, a joint study with the Municipal Corporation of Greater Mumbai was carried out on 2,000 volunteers in Inde. The participants spoke different languages, such as Hindi, Marathi, Gujarati and English. In order to increase the reliability of the test, some people were infected and others were not. Vocalis’ artificial intelligence succeeded in finding 81.2% of patients who had had a positive PCR test for Covid-19. According to the researchers, the app even successfully detected asymptomatic cases.
The start-up clarified that VocalisCheck was not a diagnostic tool but a screening tool, which can provide a reliable result. The application has obtained the regulatory CE marking, which justifies its compliance with the standards imposed for the sale of products, in theEuropean Economic Area. Currently, Vocalis only targets institutions and companies: “Employer sends email or text to employees asking them to check in before leaving home for work […] The evaluation of the vocal samples is immediate. Low-risk people are given a token that allows them to get back to work. The others are informed of the procedure to follow based on their assessment.»
The tool is available in the United States, Chile, Indonesia, South Africa, Romania and Luxembourg. For the moment, the company is not projecting itself into a mainstream application. It wants, first, to make its artificial intelligence more robust and more precise, before marketing VocalisCheck to the whole world.