Chris Mitchell makes a big noise in smart home marketplace with Audio Analytic in Cambridge
PUBLISHED: 11:16 20 May 2017 | UPDATED: 15:08 20 May 2017
Richard Marsham - RMG Photography Tel - 07798 758711
Audio Analytic is making waves in the smart home marketplace, with patented sound technology that can detect glass breaking, dogs barking, a baby crying and more. We met Chris Mitchell as the firm prepares for its move to new, larger offices in Cambridge.
There’s a baby crying, a dog is barking incessantly – and then the ‘Mario death’ sound from the Nintendo video games goes off…welcome to the strange and unique offices of Audio Analytic.
Its founder, Chris Mitchell, is a man who simply loves sound. That’s just as well. His work once took him to an RAF hangar at Alconbury to smash glass for weeks on end.
Now, as the smart home marketplace and wider consumer electronics industry wakes up to the value and potential of sound, Chris’ company is very well positioned.
As of next week, it will be very well located too, as it is moving from its home in St Andrew’s Street, Cambridge, which it has outgrown, to newly-refurbished offices in Quayside that feature unconventional shapes befitting a firm doing unique work.
Audio Analytic’s patented machine learning technology recognises sounds in their environment with unerring accuracy. The software is already in use across a range of smart home devices created by design and development companies – security products, for example, that can detect a window being smashed at your home and send an alert to your mobile phone.
“We are a machine learning/artificial intelligence company,” says Chris. “We make what we call artificial audio intelligence. That is a new academic discipline, let alone a new commercial discipline. It’s the art and science of studying sounds beyond speech and music.”
Chris’s journey into sound began with a PhD in music genre classification at Anglia Ruskin University in Cambridge.
“They basically give you a computer and some money and leave you alone for three years, which was heaven for me,” he laughs. “I was teaching machines to understand music at quite a fundamental level – like tempo, timbre and so on – and trying to stitch it back together so it could tell music was jazz or dance.
“I was self-funded and running a network consultancy business at the same time. So I’d be running around trying to fix servers at different places in Cambridge while somebody would be asking ‘When is that paper going to be ready for that conference?’ They sat in tension quite horribly! It was a question of how could I bring the two together.”
There is more than one way to analyse music.
“You can try to work out in Western music all the tempo, timbre and other attributes, but if you apply that to other types of music they might not have those attributes,” Chris explains.
“The other way is using AI, where you try to crunch the music down and present it to the computer and ask it to tell you why those sections are different from those. The technique I worked on turned out to be applicable, within reason, for any sound you wanted it to.
“The problem then was which sounds. I remember writing down a long list and trying to work out why anyone would value each sound.”
Chris won a Kaufman fellowship to go to the US, designed to help with the commercialisation of technology.
“I studied in the Bay Area for a bit and came back to the UK and did a bit of the consultancy because I’d just finished my PhD, which means your bank account is basically zero.
“Then I set about doing Audio Analytic in my spare time and weekends,” he recalled.
“We got some money out of a defence industry contract with MBDA in Stevenage to look at how our technology could be applied there. That industry is always interested in new technology. That meant I could make the leap into the company full-time.
“I found a couple of angels who gave some money along with regional development money to get it on its feet so we could file patents.
“We got a syndicate round from Cambridge Angels and got some seed money into it. We started looking at the professional security arena.”
This caught the attention of companies in the smart home marketplace which were creating domestic security applications.
“We started doing a range of sounds for them and licensing our technology out, which is our standard commercial model,” says Chris. “Then, we started working with some really cool design and product design and development companies. We were learning together and we discovered some fascinating interactions that led on to a range of commercial products.”
How does the technology differ from speech recognition?
“Speech sounds have a language component and speech recognition systems recognise a bunch of phonemes and then apply a language model to it. The two parts contribute about 50 per cent of the performance of the overall systems and there’s a huge level of detail in both.
“That fundamental work hadn’t really been done for sound recognition systems. At both of those levels we had a zero data problem. We had to go out and make the world’s largest collections of recordings for this.
“So for example, we do a glass break analytic which determines when someone is trying to break into your house. We had to collect these sounds. We had to smash, crash, break, beat and holler our way through more sounds than you probably knew existed, trying to build out this taxonomy of all sounds, which is the end goal.”
It was breaking glass that took Audio Analytic to Alconbury.
“They had an old RAF hangar there. It was huge, which meant you could fill it full of glass,” says Chris. “We stuck the glass where the jet engine was and went in a sound-proof room because this was 100-150 decibels. We spent weeks on end understanding it. We had to buy it from all sorts of places like local glaziers, then smash it up and put it in the skip for recycling. It was not a cheap endeavour!
“What you are trying to do with any given sound is understand variation, which is all the things that contribute to the individual changes in the sound. Then you can move on to the machine learning process.
“We take all the data, run it using the huge amount of algorithmic work that we’ve put in using those patented systems we’ve got and check that it can adequately classify the sound if it heard it again. Because it understands all the variation, it doesn’t matter if it hears something new – whether the glass is hit with something different or if it’s slightly bigger or thicker.”
The team had to smash different types of glass with all sorts of implements to capture variations – and consider the acoustic attributes that come into play, such as the other sounds the microphone will pick up.
“It was quite an undertaking,” Chis says. “It’s got to be very, very reliable. If it’s going to turn the lights on to scare away a burglar, or send you an alert, you don’t want to find out it’s actually the garbage men.”
The work was extended to include smoke and carbon monoxide alarm detection – so householders can be sent an alert wherever they are if an alarm sounds – and to the sound of babies crying.
“There are a whole range of products you can now buy with that technology in all over the world,” says Chris. “Whether it’s a camera, a sensor or hub, these companies are very good at making the product. Our world-class expertise is in the sound detection.
“I’ve got a one-year-old girl and I’ve got a device that turns on the hallway light low so I don’t trip when I go to her if she cries in the night.
“It’s about devices in the smart home that make your life a little bit simpler. It’s like your house is taking care of you a bit more.”
Another device enables Chris to see how much his daughter has been crying while he’s been at work – so he’s aware if his partner has had a tough day when he gets in…
“The smart home industry is about the home giving you a bit of information that can make a difference to your day,” he says.
For some, this means keeping an eye on their dogs while they’re out. About one-third of home cameras are sold for pet monitoring purposes.
“We’ve started to get into wellness and entertainment, with a dog bark sensor. We literally ran around recording dog barks,” Chris says, with a smile. “We did look at Battersea and places of that nature but we decided they were too artificial environments compared to the home.
“The engineers were having a discussion earlier about whether snout length affects the sound of a dog’s bark. There are some cutting-edge conversations that go on in our labs that nobody else is having!”
Audio Analytic first showcased the dog bark technology at the Consumer Electronics Show (CES) in Las Vegas and it can now be used in products.
It can be integrated with smart home appliances - meaning the sound of your dog barking could activate lights to deter intruders, for example, or alert consumers via their home security cameras to check on their pet.
“We are working on a whole range of sounds,” says Chris. “I sometimes lie awake at night and hear strange sounds and think: ‘That would be good to do…’ Snoring is on my roadmap!”
The firm has already worked with Intel and Cisco, along with smart home product manufacturers, and it seems likely that mobile phone firms will be interested. So what will the smart home of the future look like?
“If you take me as an early adopter, when I get near my home, my phone knows and turns the lights on and warms it up. I’ve got a bunch of home assistants – an Echo and Google Home – and I’ve got a camera in the nursery.
“But the company’s ambitions are much broader than the smart home. We’re exploring how this can impact loads of consumer electronics devices that come into contact with sounds on a daily basis.”
Audio Analytic has raised around £7million, most recently announcing investment of $5.5million at CES from Cambridge Innovation Capital, IQ Capital and Cambridge Angels.
“It’s an exciting time, particularly as audio has come to the forefront of consumer electronics,” says Chris, who says he has been helped by the Cambridge ecosystem.
“We’re currently on a course called School for Scale-ups run by Cambridge Network and we’ve done the Ignite course at Judge Business School. All the institutes around Cambridge have contributed – like IdeaSpace. So many people contributed to get it where it is,” he says.
The new Cambridge offices
Audio Analytic employs 25 currently but its new offices in Quayside can accommodate 60 – an indication of Chris Mitchell’s plans for the company’s growth. He is searching worldwide for specialist engineers and AI experts.
“We’re looking for world-class players and Cambridge is a great place to find them,” he says. “We’re looking for people who absolutely love sounds, which is the thing that holds the company together.”
And it seems to reflect Chris’s sense of humour. If someone breaks the software build in the office, it sets off the Mario death sound.
It was the culture of sound that played a part in selecting the location for the new offices. “There’s nothing richer than the sound of an urban environment. We’re a Cambridge company too. We want to be true to our identity.”
Expect Audio Analytic to make a big noise in the future then.