Artificial intelligence diagnoses cold by tone of voice

A new AI application developed by researchers at Sardar Vallabhbhai National Institute of Technology in Surat, India, can tell if the person on the phone has a cold from the voice. Designed as a resource for telediagnostics, the tool can have an impact on business.

According to the British magazine EconomistThis “cold-sounding” diagnosis that broke the news last week could theoretically unmask an employee trying to take a leave of absence, coughing dramatically as she explains to her boss the reason for her absence from work.

Delusions aside, the discoveries will be able to diagnose the common cold remotely without the need for an expensive and time-consuming doctor’s visit, according to scientists. But it doesn’t stop there: scientists from other fields are already analyzing other conditions, such as Parkinson’s disease, depression, and some head and neck cancers that affect voice patterns.

How AI can diagnose the common cold by sound

To train the AI, the researchers set off on the principle that the human voice sounds like a musical instrument, produces multiple frequencies of sound, and never reaches a completely pure note. Most of the sounds you can pick up are high harmonics. Harmonics are the sounding frequencies of a fundamental note.

By giving notes their characteristic timbre, harmonics form mathematical patterns whose frequencies are equivalent to multiples of the original note. In human speech, since harmonics tend to decrease as you move up the frequency scale, The research concluded that the common cold infection alters the pattern of this weakening.

After the researchers recorded the voices of 630 people in Germany, 111 of whom had colds, they separated the participants into wavelength spectra, making it possible to determine the dominant frequency and its harmonics. Finally, machine learning algorithms discovered patterns that can distinguish “cold” sound from healthy sounds.