Audio Analytics Released (Louroe / Sound Intelligence)

Specifically, I'd like to see how it works in group homes, halfway houses, clinics, offices, places where people can get angry and violent with little notice. Most of the places I'd use this kind of thing are government funded and unlikely to be able to afford to pay enough security officers. I'm thinking force multiplier.

I'd also like to see how it avoids false positives, say, a bunch of clients sitting in a common room watching TV. You can have a security officer or a resident supervisor sitting in the corner tweeting from their phone, but then they aren't off doing other stuff.

I wonder if it alarms on "CAN YOU DIG IT!?"

Being that I am in the NYC area and support a number of 'multi-family housing' developments, perhaps I can be of assistance in testing? I'd just need the units (and specific software?), a VMS is already in place.

Hi Mark,

We don't have any direct messaging / email capability, by design. The main concern is abuse, i.e., manufacturers pester end users. However, we could add a functionality that lets members opt-in to be contacted and/or show their email. We'll think about that.

For now, I will email the both of you and put you in touch.

How far can a microphone "hear?" A similar question might be, How far can a camera "see?" For a camera, of course, that question is dependent upon the lens that is used. Use a longer focal length lens and a camera can "see" farther away. Similarly, how far an audio surveillance system can "hear" is greatly dependent upon the microphone selected. A "shotgun" microphone, for instance, can detect sound at much greater range than an omnidirectional one. I would defer to Louroe (fine company, wonderful people, great products) as the experts in higher performance audio systems at greater ranges.

At NLSS, we took a more practical approach. We assumed the use of a commonly available microphone paired with a higher-end camera supporting AAC compression. For our tests, we used a Sony SCA-M30 microphone on an Axis camera. We assumed glass break and aggression detection would be used indoors, confirmed performance to 15 meters, then dialled it back to 10 meters in our white paper on the subject. Similarly, we assumed car alarms and gunshot detection might be used outdoors, so we confirmed 50 metres for car alarms and 100-200 meters for gunshots (more powerful cartridges detectable at longer ranges). Since the microphone and camera are paired together, we only verified that the analytic and typical camera range were a match; we did not test for the absolute greatest range where audio analytics would work by themselves in perfect conditions.

For glass break, we used a commonly available simulator at a number of trade shows including the ISC floor. For aggression, we tested it ourselves in our offices in Carlsbad, California as described earlier in this thread. Car alarm was tested it in our parking lot. Gunshot detection was tested with a variety of weapons at an indoor range in Las Vegas, as well as an outdoor range at a sheriff's department in Florida.

As with video analytics, do not expect perfect performance with audio analytics; managing customer expectations is key.

A lot of the detection ranges, especially outdoor, depend on ambient conditions. Think of the path the sound needs to take from the source to where the mic is. If there is no straight shot or line of sight, the detection range will be reduced. If the sound reflects off buildings or walls, or if wind comes into play, the range will be reduced.

That said, in most applications I doubt you'll need to go 300 feet for verbal aggression detection or even 3,000 for gunshot detection. In offices, c-stores, law enforcement facilities, etc most of the areas covered are much smaller.

I can only speak for NLSS when saying, vis-à-vis ShotSpotter, that our Gunshot Detection analytic does not provide bearing and range information; only which camera(s) heard the gunfire.

In fairness, there is an obvious admitted difference. Louroe/Sound Intel are not claiming to detect the spot of the fired shot, just that it occurred. That's lower value but also easier to do.