Researchers in the US have used artificial intelligence to produce noise-cancelling headphones that allow users to select which sounds they hear – or don't – from amplifying birdsong to blocking out the noise of a busy road.
Using deep-learning algorithms that allow users to make choices in real time, “semantic hearing” has been tested on about 20 specific noises, including sirens, a baby’s cry and a dog’s bark.
The system enables a person to programme their headphones to block out the sound of a vacuum but let them hear if someone knocks on the door.
The researchers believe that as their system can be trained to deal with any sound it could be of great use to people who suffer from sensory conditions such as misophonia, in which a noise generally considered normal, such as the clicking of a pen, can force a person from the room.
“You’re talking about augmenting human auditory perception with AI,” said Shyam Gollakota, professor of computer science and engineering at the University of Washington.
“When people hear about AI, they’re thinking about ChatGPT. Chatboxes are good if you want to make money. But I think if you’re talking about how you can augment human capability, it’s going to be really transformational for people.”
The National visited the team’s Seattle laboratory to try out the headphones and put them to the test.
Malek Itani, one of Dr Gollakota’s doctoral students and one of the system’s creators, played a succession of sounds the system had been taught to hear.
Eventually, users will have an app on a smartphone to control the sounds but for the trial a laptop programmed to block or unblock half a dozen noises was used. The headsets can also be set to cancel all noises or none.
Most startling was when Mr Itani switched on a vacuum cleaner, a sound I had blocked on the headphones, only for him to walk up behind me and knock on the desk, a nose I could hear even through the vacuum remained muted.
“There are all sort of possibilities for this,” he said with a smile.
Dr Gollakota said the project aimed to do several things simultaneously: identify all sounds in an environment and separate the target sound, while preserving its direction so users know where it is coming from – and it needs to do it within a few milliseconds of audio.
This must also be achieved without the large cloud computing usually associated with AI, but with a phone.
“We designed the first real-time neural network that can extract these target sounds and run in real time on smartphones,” Dr Gollakota said.
In November, the team revealed its findings at the Association for Computing Machinery Symposium on User Interface Software and Technology in San Francisco, California.
The team is looking to release a commercial version of the system to see the different ways their discovery might be used.
Already their work has triggered intense interest among the millions of people around the world who suffer from an auditory and sensory condition known as misophonia.
Misophonia is a disorder of decreased tolerance towards specific sounds or the stimuli associated with such sounds. These stimuli, known as “triggers”, can be unpleasant or distressing, and tend to evoke strong negative emotional, physiological and behavioural responses.
Sounds that trigger such responses include another person’s chewing and the washing of crockery.
For those who suffer most acutely, misophonia can be utterly debilitating and studies show it can affect a person’s education, professional life and relationships. Studies have linked it to a higher rate of depression and even suicide.
Cris Edwards, founder of the group soQuiet, says for some people – himself included – the sound of a person crinkling a bag or eating biscuits might be enough to drive a student from a classroom.
“It has a profound effect on a person's life,” Mr Edwards explained. “It's very common for me to be at a grocery store and if somebody is popping their gum, I have to leave and come back and shop later. It can be very isolating.”
Mr Edwards says his group has been in touch with the team at the University of Washington and believes semantic hearing could be a huge help.
“If it works as they say it does, this could be a miraculous assistive technology for people with misophonia,” he said.
Zachary Rosenthal, professor of psychiatry and behavioural sciences at Duke University in Durham, North Carolina, and a leading expert in the little-known disorder, says a forthcoming paper suggests about 5 per cent of people suffer from misophonia.
This would mean there are millions of people around the world whose lives are severely affected by noise.
Studies show the way misophonia affects a person can vary greatly, Dr Rosenthal said.
While a teenager may be “triggered” by the sounds their parents make while eating a meal, they might not have the same reaction around their friends. The reaction can also change day to day.
“Some people get triggered by some things in one situation. And then in the next situation, with the same trigger, don't get triggered as much,” says Dr Rosenthal, who is part of Duke’s Centre for Misophonia and Emotion Regulation.
Dr Rosenthal believes semantic hearing could be “transformational”, allowing sufferers to enjoy meals with their families or attend school without fear.
He gave the example of a person going to the cinema and having an intense reaction to someone behind them eating popcorn or rattling ice in a cup.
“The person would be able to stay in the movie and when they anticipate – because someone sits down behind them – they might hear crunching, they don't have to leave, they can just press the button on the app – the crunching button,” he explained.
“And now they're going to hear everything else except crunching. Game-changer.
“[This is] something that could be transformational around the world and help reduce suffering for countless millions of people – it's that big of a deal.”
