by From ShutEye to SleepScore, several smartphone apps are available if you’re trying to better understand how snoring affects your rest, allowing you to leave the microphone on all night to record your high-pitched nasal snoring and throat reverberations. However, while smartphone apps are useful for monitoring the presence of snoring, their accuracy remains an issue when applied to real-world bedrooms with extraneous noises and multiple people being heard.
Preliminary research from the University of Southampton looks at whether your snoring has a distinctive sound that could be used for identification. “How do you actually track snoring or coughing accurately?” asks Jagmohan Chauhan, an assistant professor at the university who worked on the research. Machine learning models, especially deep neural networks, may provide assistance in verifying who is performing this vocal arrangement.
While the research is fairly nascent, it’s based on peer-reviewed studies that used machine learning to verify the creators of another data-rich sound often heard piercing the pristine silence of the night: coughing.
Researchers from Google and the University of Washington combined the sound of human speech and coughs in a dataset and then used a multi-task learning approach to verify who produced a particular cough in a recording. In their study, AI performed 10 percent better than a human rater in determining who was coughing from a small group of people.
Matt Whitehill, a graduate student who worked on the cough recognition paper, questions some of the methodology behind the snoring research and believes more rigorous testing would reduce its effectiveness. However, he considers the broader concept of auditory recognition to be valid. “We showed you can do it with a cough. It seems very likely that you could do the same thing with snoring,” says Whitehill.
This audio-based part of AI is not covered as widely (and certainly not in such bombastic terms) as natural language processors like OpenAI’s ChatGPT. But regardless, some companies are finding ways that artificial intelligence could be used to analyze recordings and improve your health.
Resmonics, a Swiss company focused on detecting lung disease symptoms through artificial intelligence, has released medical software that is CE certified and is available to Swiss people through the myCough app. While the software isn’t designed to diagnose illnesses, the app can help users track how many nighttime coughs they experience and which type of cough is more prevalent. This gives users a more complete understanding of their cough patterns while deciding whether they need to see a doctor.
David Cleres, co-founder and chief technology officer of Resmonics, sees the potential for deep learning techniques to detect a particular person’s cough or snoring, but believes that major breakthroughs are needed for this part of AI research. “We learned the hard way at Resmonics that robustness to variation across recorders and locations is as hard to achieve as robustness to variation from different user populations,” Cleres writes via email. Not only is it difficult to find a dataset with a range of natural cough and snoring recordings, but it’s also difficult to predict the quality of a five-year-old iPhone’s microphone and where someone will choose to leave it at night.
So the sounds you make in bed at night can be detectable by AI and different from the night sounds made by other people in your household. Could snoring also be used as a biometric linked to you, like a fingerprint? More research is needed before we can draw any premature conclusions. “If you look at it from a health perspective, it can work,” says Chauhan. “Biometrically speaking, we can’t be sure.” Jagmohan is also interested in exploring how signal processing, without the help of machine learning models, could be used to help detect snoring.
When it comes to AI in healthcare settings, eager researchers and intrepid entrepreneurs alike continue to face the same problem: a lack of readily available quality data. The lack of diverse data to train AI can be a tangible risk for patients. For example, an algorithm used in American hospitals excluded black patients from care. Without robust datasets and careful model building, AI often performs differently in real-world conditions than in sterile practice settings.
“Everyone is really shifting to deep neural networks,” says Whitehill. This data-intensive approach further reinforces the need for bundles of recordings to produce qualitative cough and snoring research. A machine learning model that tracks when you snore or lose a lung isn’t nearly as remarkable as a chatbot that creates existential sonnets about Taco Bell’s Crunchwrap Supreme. It is still worth pursuing with vigor. While genetic AI remains on the minds of many in Silicon Valley, it would be a mistake to hit the snooze button on other AI applications and ignore their vibrant potential.
