There are those who question smart speakers for the possible privacy implications they could generate. But there are also those who see them as the perfect allies to call for help immediately when a person may suffer a cardiac arrest episode, and cannot be attended to immediately by others.
In this regard, researchers at the University of Washington have developed a tool based on Artificial Intelligence technology dedicated to the non-contact monitoring of people while they are sleeping.
For the development of this tool has been based on the monitoring of breathing through intelligent loudspeakers and from mobile phones, so that, if it detects agonizing breathing sound, it is able to ask for help immediately.
For the training, a total of 236 clips of agonizing breath sounds obtained from 162 actual 911 calls to Seattle Emergency Medical Services between 2009 and 2017 have been collected. These sounds have been offered by people who were trying to help people with cardiac arrest by bringing mobile phones close to their mouths.
In addition, recordings were also captured on different smart devices, including smart phones and smart speakers, until the 7,316 positive clips were achieved. For the negative data set, about 83 hours of audio data collected during sleep studies were used, with up to 7,305 sound samples collected.
From these data, automatic learning has been used, and under their training, this tool has been able to detect agonal breathing 97% of the time from when an intelligent device is placed up to six meters from a loudspeaker that generated the sounds. The researchers have also noticed that the tool does not detect other types of breathing, such as snoring, as agonal breathing.
With all this, the researchers have managed to incorrectly detect breathing in 0.22% of the time or less, with perfect detection in events at least 10 seconds apart.
The researchers have created the company Sound Life Sciences through which they will commercialize their technology. What is not known is whether the new technology will end up in intelligent loudspeakers in the form of skills, or will reach other types of devices that make it viable.