Audio Analytics Aggression Tested

By Ethan Ace, Published Nov 20, 2015, 12:00am EST

What if you could use your IP cameras to detect fights before they start? 

That is the goal of Louroe / Sound Intelligence with their recently released Aggression Detector audio analytics. Claiming that "90% of physical aggression are preceded by verbal aggression", these analytics are designed to alert guards before verbal altercations turn physical.

We tested the Aggression Detector using both live subjects and recorded clips of multiple scenes, to see just how these analytics perform.

Here is the app main view that we used to monitor and optimize the audio analytics:

******/***** ************ ********** ******** analytics *** * *********** useful ******** ** ************ systems, ********* ***** ** events ***** *** ********* go *********, ******** ****** fights/altercations ********. 

*******, **** *** *** suitable *** *** *********, and *** **** **** in ******** ***** *****, as ****** ** ******** ******** will ******* ******** ***** alerts, ***** **** ********** noise ****** ****** *********' ******* to *********** ******. 

Key ********

** *** *****, ******'* Aggression ********* *********** ** false *********** **** *** course ** ***** **** in ******* **** ****** area (**** ****** ***** conversations, louder ******* **** *** aisle ** ***** *******, or ***** ***********).

*******, **** ***-********** ****** patterns ****** ***** ****** to ******* *******:

  • *****/***** *****:***** ****** *** ****** triggered **** ******** ***** alerts, ********** ** ***********. Louroe ********** *** ***** aggression ********* ** ***** applications.
  • ****, ***-********** ******: ** some *****, ****** ***-********** speech, ******, ***., ********* alarms. **** *** ******** but ******** * *** times (

********* ******** ********** ******** speech ** ******* *** trigger. ***** ****** ** loud ****** ***** ********* analytics ****** *** *****.

****** ******** ***** *** as **** ** **** microphones *** **** ***********. Internal **** *** *** recommended *** ********* ****** in *** *****, **** high ********** ***** ****** drastically ******** ********* ***********. 

** ****** *** ***** integration. *** ****** ******* or ******* */* *** be ****.

*******

**** *** ******'* ********** detection ** $*** *** channel.

Potential ************

*** **** *** *** these ********** ********* ** in ******** ***** ***** where ********** ****** **** easily ****** *** ******* background *****, ********* *******, hospitals, ****** ********, **********, etc.

***** ********* ***** ** useful ** *******, *** the ********* ****** *** crowd ***** ** ****** areas ******/***** ** ** between ******* **** ****** trigger ******** ***** ******. ***** rules *** ** *** outside ** ***** **** periods, *** **** ***** largely ****** *** ******** of *** ********.

** **** ***** ***** larger ****** *** ******, such ** ********, ***** alerts *** ****** ** be **** ******** *** to ***** *****.

*******, ** **** **** areas **** ** ******* or *****, **** *** ****** unlikely ** ******** ****** speech *** ********** *** to *** ********* ***** and ********** ******, ****** the ********* **********.

************/*************

******'* ********** ******** *** a ****** ************* ******, with **** *** ******** setting: ***********, *********** ** red *****. **** ****** also ***** **** **** audio ******** ** ***** graphs:

  1. ***** ***** ********:*** *** ***** ***** audio ********, **** *** volume ***** ********* ** green, ******* ** ****** when ***** *** ********* threshold, ** ****** **** analyzed, *** ******* *** when ********** ** ********. While ********** ** ******, the ********** **** ******* from ***** ** ***** red.
  2. ***** ***********: *** ****** ***** displays ********* ********, **** higher ******* *********** ********* as *** *** **** energy, ****** *** ***, and **** *** ***.
  3. **************: *******, ** *** bottom, ******** ************** ** shown. Colors ***** **** ***** (background ***** ****) ** blue (************), **** ** green *** ***-******** ***** speech, *** ******/******/*** *** increasing ****** ** ******/**********.

** ******** ********* ********** performance *** ********* *********** over *** ****** ** multiple days ** ***** ** properly ***** ***** ******/****** activations. *********** ******** **** will **** ********* ** the ********** ****, **** settings *******, *** ********** noise ******* ** *** time ** ***********.

***** **** ***** ********, users ****** ****** **** and ***** ********. ** review ***** ******* ** more ******, ********* *** specific ***********, ** *** ***** ************ *****.

Video ***********

*** ***** ***** ******* installation *** ************* ** more ******, ********* *********** settings, *** *****, *** more:


*** ***********

******'* ********* *** *** integrate **** *** ** the ***** ******: ********, Genetec, *****, *** ********. Similar ** ***** **** applications, **** *** ** used ** ******* *** notifications ** ***** ******* in *** ******, ***** may ** ******** ********** to *** ***.

** ***** ******* *** not *********, ***** *** also ******* *****/*** ******** for ******.

Audio ******** ********

*** ******** ***** **** analytic ******** ** ******* applications, **** **** *** from ******** *****.

**** **** ****** ********** clips *** ** **** for configuration ********, ** *** frequency ******** ** *** recording ********** *** ******** playing *** **** **** are ******* *** *** impact ********* ***********. *******, ** used ***** *** ***** examples ****** ** * way ** ********** **** the ******* ** ******** changes.

**** ***** *** *****

*****, **** ******* ***** analysis ** * ***** with * ***** ** spectators. ****** ****** *** high, ***** ** *** large ******* ** ****** in *** ******** ****** and *** *********** ***** in *** ***********. ********** alerts **** ********, *** not ******** ** **** example:

****, ***-********** ******

****, **** ******* ***** loud, *** *** ********** speech. **** **** ******** occurs **** **********, ** the ******** ****** ***** all ******. *******, ******* increased ******, *** ****** is *** ******** ** stressed.

******* ***** *****

*******, **** ******* ***** stadium ***** ***** ** low ***********. **** **** lowered ****** *** ******* sensitivity, ****** *** ****** constant:

Test **********

**** *** ********* ***** an **** ***** **** Louroe ******** * *** via **-* ***** *******.

*** ********* ********/******** ******** were **** ** **** test:

  • ******/***** ************ ********** ********: 2.2.4-Louroe
  • **** *****: *.**.*.*
  • ***********: *.*.*.*****
  • ******* ******** ******: *.*.****.**
  • ********* ********: **** (*.**)

Comments (8)

Nice thorough review!

Couple questions,

Are those timelines stored somewhere, or once they move off the screen are they gone?

Did you play back the clips thru speakers into the room and bring it thru the back thru the Mic, or just bring the clips straight in thru the line level input?

Considering that broadcast audio normally has its noise floor ducked and frequency response tuned for speech, if you brought it in direct it would, IMHO, perform far better than one could expect in the real world.

I remember a while back reading that the Sound Intelligence could detect swearing. Is that part of this product? If so is it region specific, as colorful speech often has it localized variants.

Finally, you say that there were no false positives in the three days of open office deployment. But were there any false negatives? I know that when I'm troubleshooting a particularly uncooperative camera, I occasionally might indicate my dissatisfaction by sharing an expletive with the device. In rare cases I have even gone so far as to threaten components with a hot soldering iron.

So I guess I'm asking is, did you get a chance to yell in an aggressive manner, for whatever reason, and did it seem to trigger correctly?

Thanks!

Agree
Disagree
Informative
Unhelpful
Funny

The timeline images are not stored, but you can record the audio that preceded and followed an alarm.

The analytic is designed to use unfiltered audio streams and compression algorithms that don't filter out certain frequencies. So playing an mp3 over a speaker might work to trigger a detection, but isn't as ideal as a couple guys yelling at each other.

Swearing detection is an interesting idea. Certainly if you screamed a swear word you'd get a detection, but that's because of a stressed voice and not the word identification. So as long as people aren't softly whispering swear words at each other you'd still get a detection regardless of language being spoken.

Agree
Disagree
Informative
Unhelpful
Funny

I'd be curious to see if they soon further develop additional audio analytics includes keyword recognition, alerting and searching.

Agree
Disagree
Informative
Unhelpful
Funny

Does this product assume that audio is actually recorded or is it just for live monitoring? In some states audio can't be recorded legally without consent from both parties.

Agree
Disagree
Informative
Unhelpful
Funny

Louroe says: "The analytic isn't listening to words or languages but rather vocal stress and the associated change in frequencies."

That would help reduce the risk / concern. The analytic also does not record audio.

Ultimately, it is a legal issue but it does seem technically different than audio recording.

Agree
Disagree
Informative
Unhelpful
Funny

A novel use of this product would be to trigger the recording of audio only after and during detection, as such speech would not be considered interception of a private conversation, and would allow the capture of important audio details.

Agree
Disagree
Informative
Unhelpful
Funny

Thanks for the review. Regardless of test results, it is an innovative product from Louroe! You say you used an Axis Q1615 in the tests, but I don't understand where you're using the camera? Do you mean that you're just using the Q1615's audio feature?

Agree
Disagree
Informative
Unhelpful
Funny

Matt, the audio analytics app is running on the Axis camera and the mic is connected to the Axis camera. This is shown in the config video, which I've copied below:

Agree
Disagree
Informative
Unhelpful
Funny
Read this IPVM report for free.

This article is part of IPVM's 7,201 reports and 959 tests and is only available to subscribers. To get a one-time preview of our work, enter your work email to access the full article.

Already a subscriber? Login here | Join now
Loading Related Reports