Vocal Biomarkers and COVID

Research indicates that signs of infection are detectable in speech signals.

Posted Jul 17, 2020

peterschreibermedia/Shutterstock
Source: peterschreibermedia/Shutterstock

Medical researchers have mobilized like never before to combat the novel coronavirus. These efforts have focused mostly on developing vaccines or looking for new and better treatments for complications associated with COVID-19, but others have sought to design innovative methods of testing for the virus. As we wait for a safe vaccine, these new testing techniques will be vital to curtailing its spread and bringing the effective reproduction number well below 1.

One of the more ingenious approaches to testing has come out of the Massachusetts Institute of Technology Lincoln Laboratory, where a group of researchers has discovered that they can determine if a person is infected with the coronavirus by analyzing signals hidden in their speech. In a rare piece of good news, preliminary results indicate that these subtle vocal biomarkers are present among patients who have yet to develop symptoms. This could solve two immediate problems: the silent spread of the virus and the lack of real-time testing.

Problem 1: Silent Spread

Of the many “novelties” of this virus, three have made it far more transmissible:

  • One can infect others even if they are asymptomatic or pre-symptomatic.
  • An estimated 40 percent of those who have COVID-19 are asymptomatic.
  • The virus has a protracted and highly variable incubation period (2-14 days).

These three features have also made the virus exceptionally difficult to track. Not only do silent spreaders infect others, but it is also next to impossible to establish a reliable timeline that would pinpoint where and from whom they contracted the virus. Furthermore, as shelter-in-place orders have been lifted in most states, many patients may not be able to recall all the places they’ve been and every person with whom they’ve been in contact since the earliest possible time of infection.

To improve our response to the virus, we need to make testing far more common—both in terms of the number of people being tested and the frequency of the testing—and that will require reimagining how testing is currently conducted.

Problem 2: Delayed Results

At present, testing for the virus typically involves a nasal swab, which is administered by a medical professional. In the best-case scenario, results can be delivered in a matter of hours. In some cases, it may take upwards of a week to get results. Such a protracted lag period makes the results virtually meaningless.

In addition to the lag time, facilitating these tests requires a lot of infrastructures, including personnel, materials, and labs, capable of churning out results. As we have seen time and again, surges in demand have led to shortages of tests. Additional logistical difficulties, like delivering results to patients via outdated avenues like fax machines, have also been reported.

The Solution

For testing to be the most effective, it needs to address both issues (and, rather obviously, be accurate). Ideally, people would be able to take the test at their convenience in the comfort of their own homes, and the results would be available within moments. This would let them know whether or not they have become infected before they even walk outside, thereby preventing the virus from finding new hosts. Such widespread and early detection of infection would not just flatten the curve; it would cause it to nosedive.

As the saying goes, “An ounce of prevention is worth a pound of cure.”

Speech Signals

Thomas Quatieri is a senior staff member in the Lincoln Laboratory’s Human Health and Performance Systems Group at MIT, who has been researching the speech patterns of patients with amyotrophic lateral sclerosis (ALS) and Parkinson’s disease. He not only believes that these neurological diseases affect the brain’s ability to process speech, but that these changes can serve as vocal biomarkers. The human ear may not be able to discern them, but he and his team have developed a set of speech signal techniques and algorithms that can.

In some cases, they can detect these signals before the patient even knows they have the disease. Similar technology may one day be able to predict the earliest stages of psychosis with the use of nothing more than a smartphone app. One would merely have to tell the program what they ate for breakfast.

Though this may seem like a wild idea, we regularly get a “sense” of others’ moods or conditions without their explicitly telling us how they feel. This is especially true if we know them well. If a friend is feeling elated, you may be able to intuit this by reading their body language. Before they even say anything, you already know that they are coming to you with good news. Conversely, if a family member calls you and is feeling sad, you can usually tell just by the tone and tenor of their voice.

If they have a cold, the signs may be even more obvious. They may sound nasal because they have congestion; they may speak slower because they have a sore throat; they may be terse because they feel achy and tired.

In theory, other forms of respiratory inflammation should produce similar signs.

Inflammation and COVID

One of the most common complications associated with COVID-19 is acute respiratory distress syndrome, which is characterized by inflammation throughout the lungs. It is one of the most dangerous complications associated with the disease, but not all COVID-related lung inflammation reaches this level. In many cases, the inflammation may be quite mild, but it still can affect the muscle systems that we use when we speak.

Speech is an extremely intricate concert of muscle movements. It requires healthy muscles to act independently of one another. However, when one muscle becomes inflamed, it may lose some of its ability to work independently, which leads to muscle coupling. “Picture these speech subsystems as if they are the wrist and fingers of a skilled pianist; normally, the movements are independent and highly complex,” Quatieri said. If the wrist and finger movements were to become fused or coupled, this would reduce the pianist’s dexterity and force them to simplify their playing. A similar phenomenon happens when the muscles we use to speak become inflamed.

Quatieri posited that even limited coupling due to inflammation could alter the timbre of an individual’s voice enough to serve as a biomarker. This coupling would not prove that the person was infected with COVID; it would indicate the presence of inflammation, which would necessarily be present in those infected with COVID.

To test his theory, all he needed was a dataset.

“I had this ‘aha’ moment while I was watching the news,” Quatieri told MIT News. Specifically, this moment came during a segment focusing on celebrities who had tested positive for COVID. He realized that they presented him with precisely the opportunity he was looking for: Because celebrities are interviewed so frequently, he not only had voice samples from when they were healthy and when they were symptomatic but also when they were pre-symptomatic.

It did not take long for him and his colleagues to identify five subjects and find useful clips on YouTube. What they found was precisely the kind of coupling that indicated inflammation when the subjects were both symptomatic and pre-symptomatic. The coupling was absent in the samples taken before they became infected.

Though more research needs to be conducted to refine and improve testing, this is a promising item I hope we can one day put into our toolkit against coronavirus. If properly implemented, using vocal biomarkers to alert individuals to the possibility of infection could allow for more efficient quarantining, testing, and treatment from a much earlier stage. As it may be quite some time until a vaccine is developed, this kind of early detection system—one that seems to be accessible and affordable—may be our best defense against the coronavirus.