[hcfman]

I learned how to localize sound events by the differences in the sound arrival time.

And I learned how to keep sub microsecond time accuracy.

Then I put the two together and made a raspberry pi sound localizer.

https://github.com/hcfman/sbts-aru

[hcfman]

And then I got a bunch of geek friends together to cooperate with solving where the large explosions from the illegal fireworks come from. There are 5x in the group now with another couple coming online soon. Not to do anything with that except to confirm the science works over those distances, I was a kid once as well.

Anyway, the goal now is to find physical evidence that the localization was correct. In my area, there is lot of test data from the kind people that are setting off these large explosions. For that I am thankful :) Real life forensics can be fun :)

If I can find evidence that the indicated location was indeed where the explosion was set off then I will have verified that the localization can indeed be accurate over distances of almost 5km away, which is where one of the recorders is. Currently, I'm seeing a recurring pattern of explosions that appear to being set off in parking lot of a shopping area (Outside of a local Gamma in Limburg).

However, these fireworks do seem to destroy themselves pretty well, I haven't found physical evidence yet except for one that was very close (The next street), for that I found bits of the fireworks paper.

In any case, it will be quite a feat as right now the recorders are based 4.7km, 3k, > 2km sort of distances away from where they indicate that the most explosions are happening.

[jelan]

This is cool, do you think this would work as a phone application?

Does it do any kind of sound identification or matching? So if I sneeze around the device around the same time another system picks up a firework sound, will it be able to figure that out or is that a post process thing?

Would be interesting for you and your friends to pick a very exact point on GPS somewhere in the middle of all you and light off a big firework, then go back and see how close the system thought it was to the actual location. I know that kind of defeats the purpose of not launching illegal fireworks, but it’s better than shooting a gun in the air I guess.

[hcfman]

Automatic matching is the next phase. Some people have done nice Phd projects on that have have posted github code. It will take a while to integrate that because there are very few gunshots here. First I want to establish what can we ultimately expect out of it and so that's manual testing. There are several big illegal explosions a week so there's test data enough. Likely when developing an automated system I'll have to detune it or train it on fireworks just so there's something to test against.

Currently I have about 7x people in a group so soon we will be able to test if you can localize an explosion from 5km away to something the size of a small car park, that being the case you have something to justify a bigger project in a place that really needs it.

There's another, less obvious usecase as well. Currently in the US there are innocent people being arrested and thrown in jail on the basis of the sound localization and no other evidence. If the people in those areas had their own localizer they could counter false arrest and challenge localizations that could well have been the result of bounces. I like the idea of a tool that can help fight against injustice.