E-Paper
Your voice may be able to tell if you have Covid
The rush to understand, predict and head off the Covid-19 outbreak has prompted technology researchers to deploy artificial intelligence to create tools that can
The rush to understand, predict and head off the Covid-19 outbreak has prompted technology researchers to deploy artificial intelligence to create tools that can determine whether people are infected by analysing the sound of their cough, the way they speak or even breathe.
Currently, most of these efforts are at a stage in which the researchers are gathering data – speech and coughing recordings paired with information whether someone has an infection. These are then fed to AI algorithms – specifically deep learning and machine learning programmes.
“Since the voice production mechanism is so complex and dependent on cognitive abilities, any factor that affects your body or your mind will reflect in your voice. The changes can be in fractions of seconds -- what we call “micro” signatures, that are not audible to the untrained listener, but nevertheless present,” said Rita Singh, a computer sciences research professor at the Carnegie Mellon University, whose team created the COVID Voice Detector.
The changes, added Singh in a video-call, carry “a surprising amount of information that can be linked to physiological factors”. “Any condition affecting the lungs or the respiratory system – as has been established in the cases of Covid-19 infections -- has a palpable effect on voice. Moreover, the cough of a Covid patient is also distinct from a healthy person’s,” she added.
“The tool is basically a self-learning system that is now trying to learn signatures of a Covid infection that shows up in voice, not just in cough,” Singh said, adding that more data it is fed, the more it understands the signatures.
At the core of the tool is a computer programme patented by Singh and used extensively in voice profiling work by law enforcement in United States, according to a homeland security research agency.
A similar effort has been launched by researchers at the University of Cambridge. The COVID-19 Sounds app, available as a Chrome or Firefox plugin for now, is at present building a crowdsourced data set.
“Having spoken to doctors, one of the most common things they have noticed about patients with the virus is the way they catch their breath when they’re speaking, as well as a dry cough, and the intervals of their breathing patterns,” said professor Cecilia Mascolo from Cambridge’s department of computer science and technology in an article published by the university.
A third such initiative is being carried out by Mumbai-based Wadhwani Institute for Artificial Intelligence. Launched on April 7, Wadhwani’s Cough Against Covid mobile application asks users to record the sound of their coughing and share an image of a test result if they are positive.
For such efforts to work, the quality and the accuracy of data is crucial. “Such a system needs to be thoroughly vetted by the medical community and thoroughly tested. It must have near-zero false positives and zero false negatives. This is impossible to achieve in a short run, so our goal is tempered down to being able to return more conservative factoids from voice that may still be helpful to everyone,” Singh said.
As the disease brings much of the planet to a halt, scientists and health experts have called for technology to play a bigger part in tackling the pandemic – which has grown too quickly for conventional containment tools to have an effect.
Software giants such as Apple and Google are working on a mobile phone-based contact tracing tool, a technology that several nations – including India with the Aarogya Setu app – have already deployed on their own.
Apart from AI and proximity-estimating tools meant to trace contacts, researchers are also looking at internet-of-things (IOT) devices such as smart thermometers. “A company called Kinsa comes up with heat maps of fever prevalence. Such tools can be useful in identifying large outbreaks that health authorities can respond to and direct testing and containment resources if need be,” said Bhramar Mukherjee, the head of biostatistics at the University of Michigan.
The efforts also present a challenge for privacy, since connected device data, personal audio recordings, location histories and health records need to be protected from profiteering and surveillance.
India’s Aarogya Setu app has been criticised by privacy advocates for collecting too much data, the use of which is governed by privacy policies that, they say, lacks transparency and accountability. The developers and the government of India have rejected the concerns.
.
