What if the sounds we make in our sleep could predict illness days before symptoms appear? In an exclusive chat with VML Intelligence before his SXSW 2026 talk (“How Sleep Sounds & AI Detect Sickness Days Early”), Sleep Cycle CTO Mikael Kågebäck explained how his company is turning that idea into a life-saving tool. Dedicated to making healthy sleep accessible, Sleep Cycle uses its patented sound technology, AI, and data from over 3 billion analyzed nights to detect subtle audio signals. By analyzing 11 million hours of sleep audio nightly, the app aims to provide individuals and health organizations with critical early warnings—a tool Kågebäck believes could even help spot the next pandemic.
"Sound is an incredibly rich signal for health"
We already track signals like steps or biometrics for our health. Why is sound as the next big space to deepen our understanding?
Sound is nothing new. There’s a reason that the stethoscope is the symbol of healthcare. But if you compare it to the other signals, like heart rate or steps, they're super clean, they measure something very specific, so are much easier to work with. It’s only now, in the last few years, that we have been able to start really taking advantage of that incredibly rich signal that sound is, and that's all AI. We use machine learning to analyze that that signal.
What are the signals that you're trying to pick up specifically and how do you separate them from the noise?
We detect events. We listen throughout the night, and we detect inhales, exhales, your snores, and your movements. We can detect coughs and sneezes and all kinds of events.
We have been able to develop a way of robustly tracking your breathing rate and your breathing stability throughout the night. With that, we are able to do sleep staging, saying if you're in REM sleep, deep sleep, light sleep. On top of that, we have the coughing rate. By using detected coughs aggregated over areas, we can say something about how influenza is spread in the world. The last thing is measuring your own breathing rate. If you are under stress, you have a fever, or your lung capacity is less efficient because you have upper airway infections, you're going to see a jump in your breathing rate that can indicate that you are going to get sick in a day or so.
Is breathing data alone enough to detect health changes?
We are running some studies now, together with CMU and the Delphi Institute and also the UKHSA in Great Britain––– they are using our data to see if they can get earlier warnings than with other data. We have a user base of about a million people. We analyze 11 million hours of sleep every night. But you want to combine that with other signals as well. You can see that an infection is going up, but just from the cough rate, you're not going to know what kind of infection. So you want to combine that with maybe sewer samples or other data.
What is machine learning making possible that wasn't before, and what might be coming next in terms of what you can do with that data?
We are able to handle more diverse environments. Getting something to work robustly everywhere in the world, for all kinds of different people, that's difficult. Since we are available everywhere, and we have people that volunteer audio clips, we are able to collect a super diverse data set and by training on that we can make something work for everyone.
It’s only now, in the last few years, that we have been able to start really taking advantage of that incredibly rich signal that sound is, and that's all AI.
Mikael Kågebäck
CTO, Sleep Cycle
Is the intention that consumers would also be able to kind of get some kind of advanced warning that something's coming?
There's no diagnosis that we do, but we provide that data to you, and you can use that as a consumer. The second thing is, we give everyone access to Cough Radar, so anyone with the Sleep Cycle subscription can look at these coughing maps all over the world to see coughing trends. We provide warnings that coughing is rising in your area, and we give push notifications to our users so they can stay safe.
How does it work in terms of privacy?
For the Cough Radar data, we only collect the number of coughs and an approximate position and how long you slept in it. We add noise to the coordinate, and we aggregate over areas, so it will be impossible to ever track back. We keep no identifiers from night to night. It's not personal data anymore. Audio processing is processed on device.
Could this technology detect future pandemics if they were to happen?
That's the goal. That’s what we want to do with this product. I think we have a chance to make a difference.
Main image: Sleep Cycle CTO Mikael Kågebäck. Courtesy of Sleep Cycle
Please provide your contact information to continue.
Related Content
Insight
CES 2026: Neurotech
From improving sleep to upleveling gameplay, the brain has become the gateway for betterment.
Read Article 