Smart speaker skill uses AI to detect heart emergency and call 911

Posted on

When it comes to cardiac arrest, an emergency situation in which a person’s heart suddenly stops beating, the majority of cases happen outside of a hospital. This is unfortunate given the immediate response necessary from others in order to save the patient’s life. Here to potentially cut the number cardiac arrest deaths that happen during sleep is a new smart speaker skill that transforms Alexa into a vigilant guardian.

The new skill, which was created by researchers with the University of Washington, enables smart speakers and smartphones running a personal assistant like Alexa and Google Assistant to listen for the unique type of gasping sound that results from agonal breathing during a cardiac arrest. The skill was trained using recordings of real agonal breathing captured during 911 calls.

During tests, the skill was able to detect agonal breathing events from up to 20ft away with 97-percent accuracy. Though it doesn’t happen in every instance, researchers say agonal breathing is experienced in around half of all cardiac arrest cases. Of note, patients who experience agonal breathing during these cardiac events are more likely to survive.

Unlike other monitoring systems, this skill doesn’t require any sort of contact on the patient’s body, making it less intrusive and therefore more likely someone will use it. The smart speaker-based system worked even in cases where there were other sounds in addition to the breathing, such as an air conditioner running and pets making noise.

The technology is only a proof of concept at this point, one that has demonstrated yet another possible use for the technology already found in many homes. The researchers say a skill featuring this system could run locally on a device, eliminating privacy concerns related to audio uploads to a server. The team plans to launch a commercial version of the skill.

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *