From ShutEye If you’re trying to better understand how snoring affects your rest, SleepScore, several smartphone apps, are available. Microphones can be left on all night to record loud nasal grunts and gurgling throat reverberations. Its accuracy is still an issue when applied to real-world bedrooms with audible people.
A preliminary study from the University of Southampton is looking to see if your snoring has a characteristic sound that can be used to identify it. “How do you track snoring and coughing accurately in practice?” asks Jagmohan Chauhan, an assistant professor at the university who worked on the study. Machine learning models, especially Deep His neural networks, could help ascertain who is playing that snoring symphony.
This research is still in its infancy, but it builds from peer-reviewed research that uses machine learning to validate another data-rich sound maker that can often be heard through the optimistic stillness of the night. . I have a cough.
Researchers at Google and the University of Washington mixed human speech and coughing into a dataset and used a multitask learning approach to verify who made a particular cough in the recording. In their study, the AI performed 10% better than human raters in determining who coughed among a small number of people.
Matt Whitehill, a graduate student who worked on a cough identification paper, questions some of the underlying methodologies of snoring research, and believes more rigorous tests would reduce its effectiveness. believes that a broader concept of auditory discrimination is valid. “We’ve shown you can do it with coughing, and it seems very likely that you can do the same with snoring,” says Whitehill.
This voice-based segment of AI has not been covered as broadly (and certainly not an exaggeration) as natural language processors like OpenAI’s ChatGPT. Still, some companies are finding ways to use AI to analyze audio recordings and improve health.
Resmonics, a Swiss company specializing in AI-powered lung disease symptom detection, has released medical software that is CE certified and available to the Swiss public through the myCough app. While the software isn’t designed to diagnose illness, the app helps track how many times users cough during the night and what types of coughs are most prevalent. This will allow you to have a more complete understanding of your coughing patterns when deciding whether you need to see a doctor.
Resmonics co-founder and chief technology officer David Cleres sees the potential of deep learning techniques to identify specific people’s coughing and snoring, but there are still major breakthroughs in AI research in this area. I believe it is necessary. “We learned the hard way with Resmonics that robustness to variations in recording devices and locations is just as difficult to achieve as robustness to variations from different user populations,” Cleres wrote in an email. I’m here. Not only is it hard to find a dataset containing a variety of natural coughing and snoring recordings, but it’s also hard to predict the quality of his iPhone’s mic five years ago and where someone would leave it at night. is.
So the sounds you make in bed at night can be tracked by AI and may differ from the nighttime sounds made by the rest of your family. Could snoring also be used as a biometric linked to you, like your fingerprint? More research is needed before jumping to premature conclusions. “From a health standpoint, it might work,” says Chauhan. “From a biometrics standpoint, I’m not sure.” Jagmohan is also interested in exploring ways to use signal processing to aid in snore detection without the help of machine learning models. increase.
When it comes to AI in healthcare, avid researchers and intrepid entrepreneurs continue to face the same problem of lack of readily available, high-quality data. The lack of diverse data to train AI can be a glaring danger to patients. For example, algorithms used in American hospitals deprioritized care for black patients. Without robust data sets and thoughtful model building, AI often performs in real-world environments that differ from sanitized hands-on environments.
“Everyone is moving to deep neural networks,” says Whitehill. This data-intensive approach further increases the need for large volumes of audio recordings to conduct high-quality research on coughing and snoring. A machine learning model that tracks when you’re snoring or hacking your lungs isn’t as memorable as a chatbot that creates existential sonnets about Taco Bell’s Crunchwrap Supreme. It is still worth pursuing vigorously. Generative AI remains a top priority for many in Silicon Valley, but it would be a mistake to hit the snooze button on other AI applications and ignore its vibrant potential.
