How We Train Our Sensors to Recognize Sounds

Did you know that Minut can detect not only noise, but also glass break and other alarms going off in your rental property? We explain how this technology works and how we protect your privacy.
Minut Features
Calendar icon
May 27, 2022
min read
 How We Train Our Sensors to Recognize Sounds

You likely know Minut as the sensor that helps hosts and property managers protect their short-term rentals by monitoring noise and occupancy. What you might not know is that with Minut, safety goes beyond party prevention. It also includes knowing when a window breaks or the fire alarm goes off, so that you can make sure your guests and property are safe at all times. And yes, as with everything we do, these features are 100% privacy-safe!

How do Alarm recognition and Window break detection work?

They’re both powered by a technology called sound recognition.

One time we were able to come quickly because our guests had a problem with the fire alarm due to extensive baking - a host in Amsterdam

Minut can listen out for other alarms going off in your property (for example, fire or CO). When it recognizes an alarm, you will get a notification. This way, Minut makes traditional alarms smart, and enables you to act immediately should there be an emergency.

Window break detection works the same way, but for glass breaking. So you will know when a window breaks in your rental; and, no matter if it’s just an accident or something more serious like a break-in, you’ll be able to react accordingly.

That’s all you need to know to start using these features today - but if you’re curious about how they really work (in all of its techy detail) and how we ensure guest privacy, read on.

You can enable both features in your Sensor settings. Choose the sensor in the web or mobile app (by navigating to Homes, then Sensors & graphs) and simply toggle “Window break detection” and/or “Alarm recognition” on.

How does sound recognition work?

The idea behind sound recognition is simple. If humans can distinguish between the sound of an alarm going off and a window breaking, so should our sensor.

In practice, it requires machine learning and a library of sound files - not just of alarms and glass break, but also about 50 different other classes of sounds ranging from applause, cello and speech to dogs barking and vacuum cleaners. (The machine learning model needs exposure to sounds other than alarm or glass break; otherwise it would classify all sounds as belonging to one of the two categories.)

From these files, we extract features specific to each sound, and train a machine learning model to recognize those pertaining to alarm and glass break. Once trained, the model runs on the sensors themselves, which extract features locally and give them scores reflecting the probability that the sound belongs to either category. That’s how Minut can notify you about an alarm or glass break in your rental property without recording those events.

How do we keep the sound private?

Privacy and cutting-edge technology have always been at the forefront of what we do. So wanting to avoid transferring potentially sensitive files to a cloud, we implemented edge computing. 

What makes Minut different from most other similar IoT devices is that it’s actually a very small computer with an array of sensors and a powerful processor - allowing for raw sound to be analyzed locally and never leave the device. 

In other words, no sound from your home is ever sent to the cloud, nor stored there, eliminating the possibility of your guests’ data ever being exposed or compromised. Read more about how we protect guest privacy here.

What else can sound tell us?

At Minut, we’re constantly coming up with new ways to provide you with more insight from inside your rental while respecting guest privacy. We’re already using sound to detect unauthorized parties, keep you informed about alarms going off and let you know if a window breaks. Is there anything else you’d like to know about? Let us know here.

In the meantime, give Alarm recognition and Window break detection a try and let us know what you think!

A banner that says "Built for privacy" with a "Buy now" button