An algorithm can diagnose a cold from changes in someone’s voice

Faking sickness is about to get harder. Sneaking a day off work by nervously coughing down the phone to your boss might no longer cut the mustard—very soon your company might be able to tell whether or not your symptoms are real, just from the sound of your voice.

As anyone who has woken up with a runny nose and a voice like Darth Vader can attest, infection with the common cold virus alters how a person’s voice sounds. That is because the vocal cords often become inflamed, which changes their acoustic properties. The tissue temporarily swells and, therefore, vibrates at a lower pitch, which tends to make people with a cold develop a deeper voice.

A research team at the Sardar Vallabhbhai National Institute of Technology in Surat, India, tried to decode exactly how a “cold voice” differs from a healthy voice. Their research makes use of the fact that human speech, like any musical instrument, does not produce single frequencies of sounds. Even the best trained singers cannot hit pure notes like those that come from tuning forks. The dominant notes in the human voice are instead accompanied by a series of higher-pitched overtones.

Together these sets of notes fit into mathematical patterns called harmonics, with overtones having frequencies that are multiples of the original note. For example, the pitch of the second harmonic note is twice the frequency of the main note and so on. The amplitude (loudness) of these harmonics in speech tends to diminish as they proceed up the frequency scale. The team of researchers, led by Pankaj Warule, an electronic engineer, reasoned that infection with a cold might alter how this attenuation happened.

To find out, the scientists made use of an unusual resource: recordings of the voices of 630 people in Germany, 111 of whom were suffering from a cold. Each was asked to count from one to 40 and describe what they did at the weekend. They also read aloud Aesop’s fable “The North Wind and the Sun”, which has been a popular text for phonetics research since 1949.

By breaking down each person’s speech into its spectrum of component wavelengths, the researchers could identify the dominant frequency and the harmonics in each case. They then used machine-learning algorithms to analyse the relationships between the amplitudes of these harmonics and found patterns that could distinguish the cold voices from the healthy voices. They reported their work in Biomedical Signal Processing and Control.

It is not just the common cold that can be diagnosed from speech in this way. Other scientists are looking at how conditions from Parkinson’s disease and depression to head and neck cancers can also affect the patterns of frequencies in a person’s voice. It is part of a wider effort by medics, psychiatrists and computer scientists to remotely detect biomarkers for diseases in data collected on how people talk, write and even walk.

The Indian team’s diagnosis of cold voice is not yet foolproof. The results show it can correctly diagnose a cold up to around 70% of the time. Faced with another dreary Monday at the office, would you take the risk? ■

Curious about the world? To enjoy our mind-expanding science coverage, sign up to Simply Science, our weekly subscriber-only newsletter.