Cocoon: Machine learning low-frequency sound to protect your home

“Every house has its own noises, these are the ones that happen naturally, and we train our system using machine learning…” explains John Berthels, Co-founder and Head of Software at smart home protection startup Cocoon, when I meet him in London.

Launched officially on Tuesday, the company provides a small, stylish device [YouTube video] which learns which sounds are normal in your home and alerts you when it suspects something may be awry. This is coupled with occupant smartphone data to ‘see’ who is in the house.

Berthels likens a house to a fish tank where any movement at all causes ripples in the air. The system itself starts as a blank slate though, with no pre-learned information, he explains, this means it customises itself to your home and gradually learns what to expect.   

“We discussed fully disabling alerts at first,” says Berthels but “feedback from our users showed they quite like it to get in touch.” This means that while homeowners may receive alerts about noisy neighbours or Amazon deliveries over the first few weeks it quickly adapts itself and as Berthels points out: “There is no training phase for the user”. The in-device camera also provides a 30-second run up to the alert so people know what caused it.

Cocoon has been three years in the making. “We took the working proof of concept to Indiegogo,” explains Berthels, where it raised $249,002 total funds – 203% of plan – in December 2014. This “validated it was what people wanted,” he says, and so from there the company raised $4.11M in three rounds from three investors.

The initial batch of devices – all manufactured in Nottingham, UK – were shipped at the start of June. Now it retails at £299 ($400) and has orders globally which are due to be fulfilled over the coming weeks.

Berthels has a strong pedigree previously working at cloud storage startup, Humyo, which was later acquired by security firm, Trend Micro. Of the four other founders – who each come from different disciplines (covering hardware, marketing, product and commercial) – three are from Humyo.

In some ways it was a case of bringing the team back together. And this experience with Trend Micro – where Berthels went on to work for a time – helped Cocoon build security into the system from the ground up. “The problem is bad security software looks the same as good security software,” says Berthels “unless you specialise you can’t tell the difference.”

“We’re still informed generalists,” he adds “but we know just enough to be wary.” This means they made sure they hired experience in the team, took advice and once the system was built brought in an independent team of experts to rigorously test it.  

In fact, even how to get the system up and running proved a bit of a chicken and egg conundrum, he explains. The solution is, like a baby duckling, the system learns to trust its first phone via a special sound – then gets the wi-fi password and joins the home network from that first phone.

“I’ve never been involved in a project this diverse before,” says Berthels who has enjoyed bringing together cloud services, mobile apps, embedded technology, machine learning and more into one place. However, he feels the term Internet of Things is a little bit dated, from two years ago.

“The thing that gets missed it that the connections makes them a distributed system,” he says. “This is a tech wave that is less obvious that IoT.”

It is not a hard distributed system problem, he adds, but all the devices do have to collaborate. Any one element can fail at any time. “It is less about the device and more about the system”

“A device is part of your ecosystem,” he concludes. This means any device you buy has to work securely, seamlessly and instantly with your computer.


« Nutanix founder Mohit Aron talks hyperconvergence in secondary storage


C-Suite Talk Fav Tech: Oliver Tavakoli, Vectra Networks »


Do you think your smartphone is making you a workaholic?