Countless studies have established that sleep problems have multiple adverse consequences for our health, productivity at work and quality of life in general. Sleep interrupted by outside noise – for example traffic noise – is therefore a significant health problem.
In a collaboration between the Department of Computer Science, Aalborg University, the company SoundFocus and Respiratory Center Vest, Aarhus University Hospital, a solution based on the intelligent use of sound is therefore being developed to minimise the risk of being woken up by noise.
Changes in volume wake us up
The risk of being woken up is greatest when we are in the sleep phase "light sleep". The idea is therefore to register when the sleeper enters into the light sleep phase with the help of a bracelet or another wearable device. When it happens, sounds will automatically be emitted from speakers in the room, masking the outside noise and thus minimising the risk of waking up.
"It is the contrast or, if you will, the change in volume that wakes us up and not the loud sound itself. A concrete example is that we can fall asleep just fine while the television is on, but will typically wake up if someone turns it off. In the solution we are working on, we will therefore, by emitting sounds from speakers, lay out a "sound carpet" when the sleeper is in light sleep, so that the contrast with the outside noise is reduced. When the sleeper is again in a "deep sleep" phase, the sound is automatically turned off," Shagen Djanian explains. He is employed in the project as a business PhD student with the support from the Innovation Fund Denmark.
There are already several solutions on the market that use "white noise" to improve sleep, for example for babies. However, the available solutions have a number of drawbacks.
"In the solutions that are on the market right now, the sound is either constant, or it is turned off after, for example, 30 minutes. There are studies that show that it can have negative consequences for the brain to “work overtime" all night through sound exposure, so our goal is to be able to intelligently turn the sound on and off at the right times," Shagen Djanian explains.
The sound used to mask outside noise will primarily lie in the sound spectrum somewhere between "pink" and "brown noise". In addition, the speakers must be positioned correctly in relation to the noise source.
Sensor data in real time
One of the tasks in the project is to develop a method to register when the sleeper is in the light sleep phase. Initially, sensor data on heart rhythm delivered by a bracelet or other device will be used, but other physiological sensors may come into play at a later stage. One of the challenges is that today, there are very few sensors that can deliver data continuously, i.e., in real time. Typically, the data is collected and placed in the cloud, and then you cannot access it until the following day, for example.
"We are very much fans of the real-time aspect. We want to get the information continuously so that we can react to it and use it for something sensible. It is not particularly interesting to get data after the fact, and it is for sure not as interactive as in the solution that we are working on," says Kim Rishøj, Managing Director and founder of SoundFocus.
Solution on standard equipment
The solution is based on artificial intelligence, more specifically machine learning, and Shagen Djanian is currently working on developing the algorithms that will process the data to determine which sleep phase the sleeper is in. There are large amounts of sleep data available from international studies, and this data is used by Shagen Djanian to train the algorithms. Later in the project, data will be collected from trial subjects in the sleep laboratory at Respiration Center West, Aarhus University Hospital, in order for the algorithms to be fine-tuned. In a final step, the implemented solution will be tested on subjects in the laboratory.
Already today, there is advanced equipment available in sleep laboratories to measure sleep rhythms and phases. However, a prime objective of the project is to develop a solution that can run on standard consumer electronics, so that it may be used in private homes. One of the tasks in the project is therefore to find a sensor on the market that is able to deliver the necessary data in real time.
For Shagen Djanian, having the opportunity to work with something very concrete and product-oriented is an important motivational factor. For SoundFocus, the collaboration with Shagen Djanian has a clear strategic aim.
"Shagen will contribute to taking SoundFocus to the next level. Since we started the company, we have focused on solutions for the healthcare system, while the goal of this project is to develop a solution that targets the much wider consumer market. So, from the very start, employing a business PhD student has had a clear commercial objective," Kim Rishøj explains.
Over the years, SoundFocus has entered into several collaborations with research institutions, and Kim Rishøj sees this as a central element in the company's product development:
"For us, it is very important that there is solid evidence for the effect of what we develop. This is why we enter into collaborations with research institutions and, in general, focus on having more partners involved in projects, so that all necessary skills are covered.
Currently, two different types of wearable devices that provide data about the sleeper's heart rate are used in the project: A bracelet and a strip that lies under the mattress.
FACTS ABOUT THE PROJECT ”Improving Sleep Quality Using Sound Intervention”
- The project started as an offshoot of the large ISOBEL project ("Interactive Sound Zones for Better Living"), where the sleep aspect was originally included.
- There are three partners in the project: The Department of Computer Science, Aalborg University, SoundFocus ApS and Respiratory Center West, Aarhus University Hospital.
- The business PhD is financed by Innovation Foundation Denmark. See further information about the programme on the foundation’s website.
- The project runs for three years (2022-2024).
FACTS ABOUT THE TECHNOLOGY
- The sound that is emitted to minimise the effect of outside noise will not be "white noise", but primarily lie in the spectrum somewhere between "pink" and "brown noise".
- "Deep learning", a subcategory under machine learning, is used in the algorithms detecting when the sleeper is in the "light sleep" phase. The models used are "convolutional neural networks" and "recurrent neural networks".
- To train the algorithms, the data sets MESA and SHHS as well as own data are used. The training runs on Aalborg University's supercomputer CLAAUDIA.
- The solution is expected to run locally. Relatively little computing power is required to run the algorithms; a Raspberry PI, for example, will be sufficient.
- Shagen Djanian, PhD student, Department of Computer Science, Aalborg University, email@example.com
- Kim Rishøj, CEO and founder, SoundFocus ApS, firstname.lastname@example.org / 4088 7517
- Stig Andersen, Communications Officer, Department of Computer Science, Aalborg University, email@example.com / 4019 7682
Top photo: From left Shagen Djanian, PhD student, Department of Computer Science, Aalborg University, Søren H. Nielsen, Associate Professor, Department of Electrical and Computer Engineering - Signal Processing and Machine learning – Edison, Aarhus University, CTO and co-founder, SoundFocus, Kim Rishøj, CEO and founder, SoundFocus.