AI system ensures faster handling of emergency calls

Involuntary calls to the emergency centres can be handled automatically thanks to AI. This is learned from an experiment by ML2Grow at the emergency centre 112. This makes it possible to provide faster assistance and to save more human lives.

Pocket calls cost lives

Pocket calls are a known problem and daily annoyance for emergency centre operators 112.

After all, 112 is the only number that can be dialled by any smartphone for free in Europe without having to unlock the phone first. As a result, numerous smartphones unintentionally call emergency services from the pocket or through wrong touches every day. Up to 29% of the 2,660,000 calls that the emergency centres have to process every year are such unintended calls that operators lose time with unnecessarily.

Certainly during peak times, for example during a storm, when many people incorrectly call 112 for nuisance such as flooded basements, these pocket calls form an additional burden on the operators that can lead to calls being placed in a queue and it takes longer to receive assistance. turn into. After all, the emergency centres strive to answer every call within 5 seconds, which is not always possible at peak times due to unauthorized use and unintended calls. This can cost lives or cause health damage if interventions cannot be sent out on time.

A smart and efficient emergency centre

An experiment conducted by ML2Grow as part of the government innovation program Gov Buys Innovation resulted in a solution design capable of identifying the accidental calls. Based on the first seconds of a call, the system called “Hazira Digital” quickly determines whether it is an accidental or an actual intended call. Joachim van der Herten, CTO of ML2Grow, emphasizes: “When the system is insufficiently certain of the qualification as an unintended call, the call is seen as an intended call to avoid risks for the sake of certainty.” Zero risks are achieved by only switching on the system at times of nuisance and not closing the unintended calls, but placing them in a queue and giving the intended call’s priority. An actual intended call in the queue can still escape the queue more quickly by having the user push a key. “In this way, no human lives will be unnecessarily lost due to the technology, but the technology will accelerate the handling of actual emergency calls,” says van der Herten.

For privacy reasons, the experiment was carried out with barely 300 short fragments of calls and thus achieved a precision of 84%. With more training material and better training data (eg without the operator’s voice saying “hello” every time), much higher accuracy can be achieved. Within the experiment, it turned out to be impossible to build a system with existing public speech-to-text modules that, in addition to unintentional calls, also identifies unauthorized calls such as nuisance reports, although this is possible in a follow-up project.

Our technology accelerates the handling of actual emergency calls.
—  Joachim van der Herten, CTO ML2Grow

Hazira digital

The Hazira Digital system is named after Hazira, one of the operators of the Antwerp Emergency Center who gained fame in the 2016 VRT program “De Noodcentrale”, partly because of her angelic patience with which she a person with breathing difficulties on the other end of the line. knew how to recognize. It is a tribute to the many operators of the emergency centres who save many lives every day.

Datanews

Our project was featured in the Belgian ICT magazine Datanews. You can read more about it here.

Newsletter

Receive news about AI.
This field is for validation purposes and should be left unchanged.