Open Dav2070 opened 3 years ago
"COVID-19 Detection by Cough and Voice Analysis" is also mentioned in https://github.com/corona-warn-app/cwa-wishlist/issues/327
Dear @Dav2070
Thank you for your contribution. We have created a ticket for the bug Internal Tracking ID EXPOSUREAPP-5140) in the internal Jira system. Any further developments will be notified here in this Github issue.
Best wishes, DS
Corona-Warn-App Open Source Team
The problem
As discussions about reopening schools and more are starting again, it would be optimal to test every person every day. Of course, this is very difficult to achieve with the tests that are currently available.
The solution
Multiple studies show that it is possible to detect a COVID-19 infection using the recording of a cough with a very high accuracy (98,5 % of positive cases in the first study):
A possible diagnosis tool would be used like this: The user coughs into his smartphone to create a sound recording and directly gets a test result of whether the user is infected or not.
With such a tool it would be possible to test a large part of the population within seconds, as there is no logistical cost. It is completely free and takes only a few seconds to complete.
Naturally, those who get a positive result with the tool would need to do a real test for confirmation. But with such a preselection for clinical tests, the test capacity for cases with a higher probability of an infection should increase substantially.
Also, it would be possible to only let those people into public buildings like schools or to other social gatherings with a negative result of such a tool, making it much easier to track and break chains of infections and making super-spreader events much more unlikely.
The Corona-Warn-App would be perfect for such a feature, reaching millions of users. This is how I imagine this in the app:
Implementation
Diagnosis tool
A simple page in the app where the user can do a sound recording of a cough and immediately get a result, as described above.
Data collection tool
Of course, it is necessary to regularly collect new sound recordings to increase the accuracy of the ML model. With the reach of the CWA, it would be easy to collect enough data.
I think the simplest solution would be the following: When the user did a clinical test and gets the test result in the CWA (regardless of a positive or negative result), there should be a request or notification for the user to do a voluntary cough sound recording. According to the latest statistics on the website of the CWA, there were more than 8,5 million test results transmitted to users. Even if only a few percent of those users were doing the voluntary sound recording donation, there would be plenty of data to train the detection model.
Initial development
There are multiple projects which collected sound data in the last months. For example, a project by the University of Cambrigde collected more than 65.000 sound recordings from 32.000 participants. They are willing to share the data, so this could be used to build an initial version of a diagnosis tool.
If done right, such a simple feature could change the course of the pandemic, as rapid testing could be done by anyone within seconds and with no cost. But this technology is only effective at a large scale, as the CWA has. Now would be the perfect time to implement this as we need rapid testing to make reopening the society possible.
Internal Tracking ID EXPOSUREAPP-5140