Google AI tool could potentially be used to diagnose a person's cough

The technology system called Health Acoustic Representations (HeAR) was trained on millions of audio clips of human sounds from the YouTube database.

Google scientists are zoning in on health care products and innovations powered by artificial intelligence, including a tool that could detect a health condition based on audio of coughing and breathing. 

The AI system called Health Acoustic Representations (HeAR) was trained on millions of audio clips of human sounds from the YouTube database and could one day be used by physicians to diagnose diseases or gauge the function of a person’s lungs, according to the science and technology journal Nature. 

The journal said using sound as a biomarker for diseases was a concept born during the pandemic when research groups found it possible to detect COVID-19 through the audio of a person’s cough. But there is currently no tool approved by the Food and Drug Administration to diagnose different health conditions using sound. 

What’s unique about Google’s HeAR technology is how it can be fine-tuned to perform multiple tasks through something called supervised learning, Nature said. 

The Google research team converted over 300 million sound clips of coughing, throat-clearing, laughing and other human noises into spectrograms, which are visual representations of audio frequencies. By blocking out certain segments of the spectrograms, the researchers were able to train HeAR to predict the missing portions.