Breaking the AI sound barrier

Building artificial audio intelligence, one glass smash at a time.

Breaking the AI sound barrier

5 years ago
Duration 0:46
Breaking the AI sound barrier

If you're driving north of Cambridge, U.K. and hear the sound of breaking glass coming from an airport hangar, it's not vandalism.

The noise is being made by Chris Mitchell and his team at Audio Analytic and they believe they're breaking the AI sound barrier.
Breaking the sound barrier? A member of the Audio Analytic's team gathers the sound of glass breaking for the company's sound recognition process.

Chris is the founder and CEO of the company. They're working on something they've dubbed "artificial audio intelligence," which he describes as a new type of sound recognition for smart home applications.

"What it does," Chris says, "is discover the underlying language of sounds and enables machines to then detect those sounds and respond to them."

This area of AI is fairly new, because Chris says sound recognition is a zero data problem: "If you haven't got any data readily available obviously there's not very much to teach an artificial intelligence machine."

Audio Analytic is working to develop a database of sounds mainly related to home safety and security. Some of those sounds include baby cries, dog barks, smoke detector alarms, and breaking glass.

Gathering the sound of breaking glass was a particularly huge undertaking which involved setting up at an old RAF hangar.
Dr. Chris Mitchell is the founder and CEO of Audio Analytic. His PhD is in sound information systems and signal processing.

"It's designed for testing jet engines in it which is perfect for us because the bit where the jet engine goes is very soundproof and then you've got a huge hangar that you can put glass inside of," says Chris.

The team then spent weeks, often wearing protective Kevlar gear, breaking different types of glass and recording the sound. 

All those authentic sounds being gathered by Audio Analytic are crucial for accurate sound recognition.

"If you go onto something like Google or YouTube and you type in 'glass break recordings', you'll often get back what Hollywood thinks glass break sounds like. In actuality the sounds are very very different and a lot more diverse," Chris says,

"And if you trained a system off that sort of YouTube type data it would have severe limitations and wouldn't work in the real world."   


To encourage thoughtful and respectful conversations, first and last names will appear with each submission to CBC/Radio-Canada's online communities (except in children and youth-oriented communities). Pseudonyms will no longer be permitted.

By submitting a comment, you accept that CBC has the right to reproduce and publish that comment in whole or in part, in any manner CBC chooses. Please note that CBC does not endorse the opinions expressed in comments. Comments on this story are moderated according to our Submission Guidelines. Comments are welcome while open. We reserve the right to close comments at any time.

Become a CBC Member

Join the conversation  Create account

Already have an account?