The Case for Sound Analytics

agression detection

It’s not new — a company in Netherlands has been working diligently on this for at least 20 years. It is a company you’ll want to hear all about, however, and it is a company all about hearing: Sound Intelligence.

The company is a global leader in advanced audio analytics. Based upon original research at the University of Groningen, Sound Intelligence built and maintains a gigantic database of sounds. More importantly, the company developed deep learning algorithms to detect specific sounds and high-performance applications that put that software to use — patented technology based on the operation of the human ear.

There are noises from the school parking lot: Is it teenagers having fun or a fight? Sound Intelligence detects and reports the raised volumes, any aggression in the voices so video can swoop in for a confirming look.

If, in security, video is Batman, then audio is Robin. And the whole world is Gotham City, full of arresting sounds.

Given modern machine-learning techniques, the number of cataloged sounds in a world-class analytics library are massive, the biggest of Big Data. It has taken decades to accumulate the Sound Intelligence database. While that sounds like slow-going, there’s a tipping point where the database is so enormous (and therefore more definitive) that it’s a barrier to entry for would-be competitors.

In a security context, some of the most common sound signatures which analytics listen for include aggression, car alarms, gunshots and breaking glass.

gunshot test

Sound Intelligence often refers to daytime versus nighttime scenarios, not unlike how video analytics view the challenges of day and night as different. Most of us are familiar with video analytics: They enable cameras daily and nightly to measure crowd size, alert us to strange behavior or detect intruders.

Unlike a video security camera, audio analytics do not continuously record but only operate in buffer mode, saving the sound a few seconds before and after the detection (allowing security to verify the sound and preserve it for evidence). Sound detection does not use or record any “spoken” words for detection, only the frequency, pitch, and other characteristics of a voice or a sound event. Audio detection is NOT a recording exercise, but an alert system.

Sound intelligence software needs to listen over time for a complex combination of characteristics — from decibel level to different frequencies to the energy in those frequencies — while ignoring background noise such as machine noise, traffic, conversations, music and more.

Today these applications can trigger an alert whenever they hear a specific acoustic pattern and yet minimize the risk of false positives/negatives even in challenging environments (like noisy hospitals, train platforms and prison cellblocks.)

Sound analytics are best used to augment video surveillance systems. Sound detection enables security to be faster and more proactive in identifying and responding to threats. Sound Intelligence sells only via integrators, companies which usually serve verticals such as health, education, transport, corporate or retail (including retail banking).

While the technology today is more prevalent in use in larger organizations, the day will come when companies of all sizes will add this to their security profile. After hearing about this integration opportunity for the built environment, the case for sound analytics seems to ring true—if ProAV integrators are ready to listen.