We all know what a person’s voice sounds like when they have bad cold. But would you recognise the sound of some with Parkinson’s disease? What about dementia or neck cancer?
Probably not, but researchers believe algorithms and artificial intelligence will soon be able to pick out signs of illness from someone’s voice.
The technology could even be useful for diagnosing the outbreaks of the coronavirus.
“Each one of our internal organs is sort of a resonator, so if we have a problem with our lungs or our heart … this is reflected in our voice,” says Giovanni Saggio, an electronics engineering professor at Rome’s Tor Vergata University, researching in this area.
“The way our voice sounds also depends on who is directing the music, that is, the brain. If you have a cerebral problem, like Alzheimer’s or Parkinson’s, the way you talk changes, and we can detect that.”
The current novel coronavirus epidemic could be picked up this way, Saggio suggests, since the virus “compromises the lungs and airwaves, so the voice is definitely affected.”
Suspect cases could be screened remotely, keeping them away from hospitals and reducing contagion risks, Saggio says, adding that he is trying to contact Italian hospitals dealing with the outbreak.
He’s one of several researchers around the world looking into this topic. In Germany, a spin-off from Humboldt University, PeakProfiling, says it can “build algorithms which detect emotional states and physical conditions purely from sound data – especially from the human voice.”
Meanwhile, University of Washington researchers have published a study in Nature on how devices like Amazon’s Alexa can successfully recognize the gasping sounds that often pre-announce cardiac arrest.
By Saggio’s reckoning, more than 150 scientific papers have been published worldwide on the correlation between voice changes and various pathologies.
Using his own patented technology, Saggio tested 284 tuberculosis patients in Mumbai, measuring them against a healthy control group of 28, asking each person to record the same short sentences.
When voices were analysed, differences in acoustic parameters revealed who was healthy and who was not. Parkinson’s sufferers were correctly detected in 95 per cent of cases.
There may be even more applications, as Saggio imagines that voice recognition could help a parent understand why their baby is coughing or crying.
Saggio recalls there was even a diagnosis within his initial trials. “There was this man in the control group who was meant to be healthy,” he says.
“But we told him: Looking at your data, we think you might have a temperature tomorrow, and he said, ‘But I feel well!’ The next day he had a temperature.”
Saggio had his study published in the Journal of Communication, Navigation, Sensing and Services in 2016 and is promoting his invention through a start-up, VoiceWise.
He has been working on it for around 10 years, and, aged 55, hopes to see its wider adoption before reaching retirement age. “It’s going to be a matter of a few years,” he says.
The technology could deliver significant public health benefits, as it would make it easier and cheaper to screen patients, including in developing countries, resulting in earlier diagnoses.