Vietnamese engineers are encouraging the community to submit 5,000 different cough recordings. This will be used as data for building AI software — AICOVIDVN — that can detect COVID-19 based on how the cough sounds.
Khoa Hoc Phat Trien reports that the AICOVIDVN project is a joint force between 200 specialists and over 1,000 engineers locally. The concept of the product is simple: users will cough into the microphone of their smartphone, and the AI system will analyze the sound and give a diagnosis of possible COVID-19 infection. This will hopefully ease the process of identifying positive cases as it is easier, faster and more convenient on a wider scale.
Currently, the team has processed and labeled over 7,000 samples of coughing audio data received from other countries. Now, they need at least 10,000 data points from Vietnamese people in order to make the software more accurate. The project is seeking 5,000 more submissions of cough recordings.
The biological characteristics that are being studied from the coughs are:
- Changes in the sound produced by vocal cords
- Changes in the mood and expression
- Changes of the sound from the lung and airways
- Muscle degeneration
Through these analyses, the AI will be able to detect the effects on coughs caused by COVID-19 and identify people who have the disease from people who do not. This is something that is impossible to do with just the ear. Besides Vietnam, similar AI models are being researched in India, the United Kingdom and Switzerland.
The engineers of AICOVIDVN shared with Khoa Hoc Phat Trien: “Because the coughs of people from different countries have different characteristics, we need the cough of Vietnamese people the most.”
The project is receiving recordings via Zalo and Messenger. All Vietnamese people, with priority for F0, F1, F2 and F3 cases, can make submissions through the following steps:
- Start the recording and cough four to five times.
- Add a message: “I believe I am F…” (Skip if not applicable).
- Send.
[Top image via Cambridge University]
