Hey, Soldier: Analyzing Body Camera Audio to Improve Police Tactics

Tejas Shastry

October 30, 2020

Blog

A friend of mine is now an agent in Homeland Security, but he started his days off in the Chicago Police Department as a beat cop. He mostly focused on underprivileged neighborhoods in the city and would go around the blocks, do his patrol, and try his best to carry a friendly demeanor. But, after a few weeks on the job, he just couldn’t seem to get any respect or positive response from anyone in the neighborhood.

He went to one of the more senior cops he worked with and got a little piece of advice — if you want to be effective with young guys on the street who are involved with gangs or dealing drugs, you have to acknowledge the world they come from. His colleague gave him this little tip — next time you say hi, try calling them soldier. It acknowledges their experience in the world.

So he did. And just that simple change in language helped him build relationships with that community.

It should come as no surprise to anyone listening that policing in America and across the world is under the magnifying glass. Tactics like the one I described are just one example of a small change police interaction that could help save lives across the board.

So, we asked ourselves at GreenKey, how can we play a role in helping to improve policing? How can our technology help surface these ideas?

At GreenKey, we’ve developed cutting-edge natural language processing that can transcribe audio, identify sentiment, tag topics, and more. We got our start 7 years ago helping Wall Street banks analyze their conversations to identify missed opportunities and reduce fraud. Over the last 3 years, we’ve worked with public safety telephony providers like Motorola Solutions to transcribe 911 calls for emergency dispatchers.

And now, we’re using our natural language platform to help departments improve their policing with actionable data.

Building Models to Identify Tactics

If you’ve ever listened to a body cam video, you’ll know that the audio track is a mess. It’s noisy, full of multiple speakers, and (no surprise) challenging for machines to transcribe. Body cam transcription can provide valuable evidence for criminal investigations, but accuracy often holds back its value.

However, in order for a machine to understand what’s happening within a body cam audio track, it doesn’t need 100% accurate transcription. Similar to a human being, GreenKey’s natural language processing platform can understand if a person is expressing negative sentiment, is nervous or not confident, or is trying to de-escalate a situation, all without having 100% accurate transcription of the words being spoken.

Our Focus Studio product enables crime analysts to analyze data, tag relevant events and sentiment, and build models off of our base model that can help surface tactics and issues in body cam audio.

Understanding what works and what doesn’t

Once GreenKey’s engine has structured body cam audio, it becomes a powerful datasource to start analyzing police tactics. Our engine extracts transcription, intent, and sentiment side by side, giving a detailed view of events as they happen. In the example below, it’s clear that a suspect is being aggressive, leading to negative sentiment. De-escalation tactics by the officer seem to provide some relief, but eventually sentiment turns sour again and the suspect must be arrested. 

Seeing data like this is like turning on the lights for many departments. It’s challenging to improve training and oversight when you have little visibility into what tactics actually work. By leveraging natural language processing, cities and police departments can finally get quantitative data on ways to help their officers improve police tactics, ultimately saving lives across the board.

This is brand new technology and we’re just starting our first engagements with pilot customers. If you’re a department that’s interested in unlocking your body cam data or a crime analyst interetsed in building a model with GreenKey’s platform, let us know.