People are finding new ways to use artificial intelligence in a medical environment. Now, researchers from India have developed an AI tool to detect if someone has a cold just by reviewing their voice tone. This could be used to aid diagnosis and even to identify if an employee is taking a fake sick day because of having a cold.

A team at the Sardar Vallabhbhai National Institute of Technology (SVNIT) developed an algorithm that can distinguish between a normal voice and one with a common cold with just one or two words. It’s close to 70% accurate and the idea behind it was to develop speech signal-based non-invasive diagnostic techniques that can work remotely.

“We studied a database of voice samples of individuals and extracted features from the voices that had cold or those that were normal. Then the machine was trained to identify the features through deep learning,” Suman Deb, study author, told The Times of India. “One or two words are enough to confirm if the person has a common cold.”

Looking into voice features

The common cold is a viral infectious disease of the upper respiratory tract. A person suffering from a common cold can show various symptoms such as coughing, sore throat, runny nose, sneezing, and hoarseness. Cold speech has a lower pitch, an increase in noise due to coughing or hoarseness, and also a change in the timber of the voice.

In their study, the researchers made use of the fact that human speech doesn’t produce single frequencies of sounds, as reported by The Economist. Rather than having just one leading note, speech is accompanied by a series of overtones higher in pitch. These follow mathematical patterns called harmonics, where their frequencies are multiples of the initial note.

The team hypothesized that a cold infection could modify the pattern of attenuation in the amplitude, or loudness, of harmonics present in speech as they ascend the frequency spectrum. Typically, the higher-pitched harmonics exhibit a decrease in amplitude as their frequency increases. To find out, they took advantage of a unique resource.

They studied voice samples of 630 people in Germany, 111 of whom were suffering from a cold. Each had to count from one to 40, tell what they did at the weekend and read from a popular text for phonetics. The researchers used machine-learning algorithms with the recordings and found patterns that differentiate the cold voices from the healthy ones.

While impressive, this isn’t the first time artificial intelligence has been used for medical purposes. AI has been recently used to diagnose heart disease (by looking at eye scans), pneumonia (by listening to coughs), and even rare genetic diseases (by looking at photos of a person). In another study, researchers created a method to find high-affinity antibody therapeutics based on AI, which could accelerate the discovery of novel drugs. Another group of researchers has also used AI to detect Parkinson’s through breathing patterns. The accuracy can still be better and AIs won’t replace doctors anytime soon, but they could serve as tools to aid doctors’ diagnosis.

The study was published in the journal Biomedical Signal Processing and Control.