London 5th September : An app for smartphones that can pinpoint Covid-19 virus in voice recordings of individuals with the aid of artificial intelligence (AI) researchers have revealed on Monday.The team claims that their app is more reliable than other antigen tests, and is also cheap, fast and simple to use, which means it is suitable for use in poorer countries, in which PCR tests are expensive and/or difficult be distributed.
“The positive results indicate that voice recordings made with a simple microphone and tuned AI algorithms could be able to achieve high-precision in the identification of patients with Covid-19 infections,” said Wafaa Aljbawi researcher from the Institute of Data Science, Maastricht University, The Netherlands.
“Moreover they allow remote virtual testing and offer a turnaround time of less than one minute.They can be utilized for instance, at entry points to large gatherings, and allow for quick screening of the crowd,” she said at the European Respiratory Society International Congress in Barcelona, Spain.
Covid-19 is a common condition that affects the vocal cords, causing changes in the voice of a person.
Aljbawi and her supervisors decided to look into whether it was possible to make use of AI to analyze voices in order to find Covid-19.
They utilized data from the University of Cambridge’s crowdsourced Covid-19 Sounds App which contains 893 audio samples of healthy and unhealthy people 308 of who had tested positive for Covid-19.
The researchers employed the technique of voice analysis known as Mel-spectrogram analysis.This technique detects various voice characteristics like power, loudness and variations over time.
“In order to differentiate the voices of covid-19 patients from those without the illness, we developed various artificial intelligence models and compared which one performed the best in classifying cases of Covid-19,” Aljbawi added.
They discovered that one model, called Long-Short-Term Memory (LSTM) outperformed the other models.
The LSTM model is built on neural networks that mimic the way in which the human brain functions and can recognize the connections in data.
The overall precision was 89 percent Its ability to accurately identify positive cases was 89 percent and its ability to correctly recognize negative cases was 83 percent.
“These results demonstrate an impressive increase in the precision of diagnosing Covid-19 when compared to the most current tests, like the test for lateral flow” Aljbawi added.
The researchers state that their findings must be verified with huge numbers.
na/uk
