Let me follow up, and maybe I should write this up on CAB but I don't really want to get too much into police reform there, what I know factually about the ShotSpotter system. Most of this comes from discussions with APD leadership and officers in the context of the police oversight role I used to hold, and they are very much limited in what they can say. Some of this is specific to APD SOP and other police departments may vary in their approach.
Some sort of software analysis performed by SoundThinking identifies a possible gunshot. The audio recordings are sent to a human analyst, a SoundThinking employee, who reviews the recording and enters an assessment of what it contains (e.g. if it is gunfire, and how many rounds). If the analyst confirms the report, an alert is sent as a text message (I believe in an app they furnish) to staff in APD's dispatch center, called the Emergency Communications Center (ECC). The contract includes an SLA on this process of I think 1 or 2 minutes, but I was told that they routinely performed as well as 30 seconds. Some APD personnel, I think usually area commanders but it may have been all field division sergeants, also receive the alerts on their phones.
The ECC dispatches the call as a priority 2. P2 is high enough that a ShotSpotter report will "bump" most calls for service except for a caller on the line violent crime in progress. When the officers arrive at the reported location, they make a brief assessment and search the area for suspects or victims. If no suspects or victims are found, a Crime Scene Technician is dispatched (often later as they will wait for daylight) to search the area for evidence such as spent shell casings.
My recollection is that I was told they were able to find definitive evidence of actual shots fired in less than five percent of cases, but take that with a grain of salt as I do not believe it was ever put in writing (I don't think they're allowed to by their contract) and I could be remembering wrong. However, it's believable that the accuracy of the system is higher than that suggests, as Albuquerque has a lot of wide open spaces that are difficult to search thoroughly if the ShotSpotter location estimate is at all inaccurate.
I was told that, when the system was tested by firing blanks, it was not completely effective but that they were satisfied with how effective it was. I was never given a number and I think they had been very specifically prohibited from discussing the testing in detail when they coordinated it with SoundThinking.
One of the major criticisms of the ShotSpotter system in Albuquerque is that it results in a relatively large volume of P2 calls that delay police response to most other calls for service. During the worst of the understaffing, I was told that some officers in high-crime areas like the International District spent a large portion of their total shift following up on ShotSpotter activations while there were multiple P3 calls queued. This has probably improved as staffing levels have increased, but in my mind it is the greatest single concern about the system.
SoundThinking's evasiveness, refusal of independent research, and clear motive to sell their product creates an alarming possibility that they are deceiving police leadership and elected officials into overprioritizing ShotSpotter. It may be a waste of money, which is already a problem, but the much greater concern is that police departments are putting off responding to nonviolent crimes in progress to go to ShotSpotter reports instead, because they have been told by SoundThinking that the accuracy rate of the system is very high.
and with manual input you can greatly improve the accuracy from what an automatic system could do due to noise and signal degradation. However it’s very time consuming. The service level response would not be met if doing this on many cases I’m sure. Meaning the default accuracy of localization is likely to be a bunch less accurate than what it could be in theory.
What is absolutely needed is to start a validation process. Start legal proceedings to force the disclosure of co-ordinates and exact event time information as recorded by each recorder so the math can be checked. My software will provide that side.
This looks awesome. I’ve been wondering if it would be practical/feasible to create something similar but portable, self-contained for quieter, „local” sounds, like for instance locating a termite-like bug(s?) that’s been terrorizing my sanity for years now hiding in wooden beams.
Yeah, for short close distances sound camera of phase difference of arrival localization is interesting.
I did wonder about finding mice. I connected an ultra-sonic microphone once and saw lots of squeaks that were probably mice. I suspect the accuracy is sufficient.
Previously done a lot of research into firearm and explosive acoustics
(and the DSP to locate and categorise) and can say this is a _hard_
problem. Military versions for use in wide open spaces have a much
better chance than in urban areas.
Less than five percent seems unlikely, but I wouldn't be surprised to
hear "less than half the time".
Categorisation will certainly be error prone as there is so much similarity because fireworks and gunshots. The main difference I expect is the speed of the shockwave resulting in steeper spectrograms for gunshots. However, a steep spectrogram is essentially showing a large amount of high as well as low frequencies. The higher frequencies degrade very quickly over distance reducing the difference of a gunshot from a firework the more this degrades.
Not finding evidence is a different problem, it could be as simple as using a revolver rather than a semi-auto that ejects the cartridges.
In any case, it would be great to see independent testing of this problem by the sort of people in this forum and you can use the software I developed last year (https://github.com/hcfman/sbts-aru) to do so. That software sets the time on the Raspberry Pi to have less than 1 microsecond of error, which is more than enough for any validation efforts.
Nice work. PM me if you want to chat. Have you considered the
possibilities with oversampling and adding more channels- now there are
much cheaper 192kHz multi-chan ADCs on the market - these can be used
with MEMS arrays. Then you can play with phase information much more
freely.
The frustrating problem with people down voting without any
explanation is that simply ignoring the knowledge of experts and
hiding whatever has emotionally enraged you does not further the
discussion.
I think people don't "trust the experts" when they read the "experts" saying things like this:
> The reader can probably infer how this coverage pattern relates to race and class in Albuquerque. It's not perfect, but the distance from your house to a ShotSpotter sensor correlates fairly well with your household income. The wealthier you are, the less surveilled you are.
This sets off the bs alarms to me. The author knows this is a gross distortion of the rationale and is playing dumb. So what else are they lying about?
1. Cops use enforcement on poor neighborhoods because they like to make their lives difficult and because they hate poor people
2. Cops use enforcement there because that's where the crimes take place
Explanation 2 is so obviously much more likely that I can't take the person that purports reason #1 without acknowledging the other explanation. They're playing dumb.
1) Why is this conversation happening in this subthread? What does this have to do with nonrandomstring's experience in doing similar work for the military?
2) Nothing in the passage you quoted insinuates anything about the reasoning for surveiling poor neighborhoods more than rich. It simply points out that this is a thing that is happening.
Earlier they said this:
> Many assumed that ShotSpotter coverage was concentrated in disadvantaged parts of the city, an unsurprising outcome but one that could contribute to systemic overpolicing.
It's unsurprising because that is where the crime tends to take place, but there's a valid concern raised that this is a self-fulfilling prophecy.
I've read some highly political anti-police rants in the past few years, but this isn't one of them. The author barely makes their opinions known at all, and when they do they're very aware of the complexities.
So the bodies of people shot are seen more often in high police presence areas and just not found in other areas? That seem strange that it would be due to higher police presence. The fact that for non violent crimes the stats are skewed sound plausible but that seem less so for violent ones.
Higher enforcement of nonviolent crime leads to more strained relationships with the community leads to less cooperation with law enforcement leads to higher rates of all crime.
3. Cops use performative enforcement in poor neighborhoods because it yields metrics important to them with minimal effort.
Poor neighborhoods are often demanding and crying out for enforcement of laws. Nobody wants to live next to a crackhouse, and the people forced to lack options to leave. In my city, there's a notorious drug house that has been in operation since I was in college in the 1990s. It has better staying power than Walmart -- it's still there. What poor people in general resent to the point of riot about is systematic bullshit and abuse on the part of ineffective police.
When the powers that be say "clean up that area", they round up stupid kids for bullshit, maybe hit up a few street dealers, etc. The actual hard work, say arresting street gang leadership or investigating property crimes isn't sexy and takes time. The best documented examples are NYPD, where the worship of Excel sheets resulted in sweeps where the cops would issue appearance tickets for such offenses as "obstructing the public sidewalk" (ironically doing so, btw, when they are literally parking their personal vehicles on sidewalks for these big sweeps), than run through and make arrests for failure to appear a few months later.
Shotspotter in particular is stupid - it's just a way to blow Federal grant money. If as you say, the police "know where the crime is", why would they need microphones to tell them where gunshots go off? Presumable they patrol high risk areas and hear it themselves.
Police departments are paramilitary organizations. That means they need a military like level of accountability and discipline to function well, and the nature of modern governance is such that that is lacking. The Army doesn't tolerate drunken soliders runing amok, but police departments do. IMO the best way to address the issues of policing is to consolidate smaller departments into state or regional entities to both professionalize and reduce the chummy nature of what goes on.
> When the powers that be say "clean up that area", they round up stupid kids for bullshit, maybe hit up a few street dealers, etc. The actual hard work, say arresting street gang leadership or investigating property crimes isn't sexy and takes time.
Shotspotter is useless (5% is at best par for the course for every intelligence appliance I've used myself) but your expectations are unrealistic.
You're berating them for shaking down street kids for intelligence, then berating them for not taking down gang leaders? Lmao.
That's an impossible situation you've created. How about you demonstrate how policing should be done?
> why would they need microphones to tell them where gunshots go off?
Half my family are police. It’s impossible to do the actual work. Half of them plan to retire at the minimum years of service and transition to private security to a fire department.
Writing tickets for blocking sidewalks is an overtime generating detail, not intelligence. It’s “objective” in the sense that everyone gets harassed, and thus safer than using discretion to do something potentially intelligent.
Of course number 2 is the reason. The author knows that too, it's obvious to the point of not needing mentioned. It doesn't change what they say though, the end result is that poorer people are more surveiled.
Ah. Not what I expected. And that's really why it ought to be
necessary to leave a comment in order to make a downvote.
I assumed someone was raising a technical objection to my claim that
acoustic location of transient sounds is "hard" - myself being the
"expert" having published on firearm signatures and machine listening.
But this is about the socio-politics in TFA. Okay.
The quote doesn't insinuate anything like what you claim it's insinuating. I don't believe you're intentionally distorting their intent, but I do think they you're reacting to other pieces you've read before in a similar genre rather than to the article and author at hand.
Wow that's even worse than what I had read from reports about the deployment in Chicago (which I read up on when Seattle was considering it). I think the value was like 10%.
Here's a decent article because this happened in recent weeks (1). The numbers they cite there are 89% of alerts lead to no reports of gun crime and 86% lead to no crime reports at all.
All that being said, the metrics are horribly flawed because they assume police will actually investigate anything after dispatch and that the investigation will turn anything up. Believing that CPD is going to case a neighborhood looking for GSR and shell casings with a forensics team is a tall order. If there isn't someone bleeding out in the street there probably isn't going to be a police report filed.
And even that said - ShotSpotter is useless because it requires police to do their job to be useful. When Reddit is better at tracking gang violence than cops, no fancy audio forensics tool is going to help.
"Definitive evidence of shots fired" is a higher bar than shots actually being fired. I agree it's still very low though, especially since it means that at least 95% of the time responding to the call was a complete waste of time (= even if shots were fired, there was no evidence of crime).
My read is that they know it does work but they don’t want to continue to use it because statistically the “wrong” people are committing the crimes it’s tipping off.
I mean brass can be crazy hard to find, I'd bet if you did a geocaching experiment with brass picking a random spot within shotspotter's accuracy tolerances, you'd find the casing less than 50% of the time. Add in an hourly employee that is in no way motivated to find the casing and you'd prob drop to 10%.
Though at the end of the day it doesn't really matter. The brass I mean. Finding brass or not finding brass doesn't exactly help catch or prosecute someone shooting a gun illegally. The whole point was to more quickly catch crimes in progress, and it doesn't seem like that's working.
Do you know how it communicates with ECC? Can it be MITMed?
Would it report if you were playing Counterstrike outside with speakers on? What if you were listening to some gangster rap with shots in it? Or just a Raspberry Pi with a speaker on with gunshot? It seems to me something like this can in theory DoS the ECC.
All of these are much more complicated and expensive than just sending someone off in a stolen car to fire some shots wherever you'd like the cops to be in 20 minutes or so. Convenient leash.
I'm sure there are ways to make a gun shot sound without having an actual gun, too. If police responds every time within a few minutes, it's also very easy to find out which device/firework will do.
Depends on who's doing it. For career criminals, sure the stolen car and gun are probably easy. If it's anybody else, say some bored high school kids, then the speaker is probably a much easier option.
Yep. I guess using a gun didn't cross my mind as possession is illegal here in this part of Europe (unless you are licensed, screened, store it safely, etc etc).
Some sort of software analysis performed by SoundThinking identifies a possible gunshot. The audio recordings are sent to a human analyst, a SoundThinking employee, who reviews the recording and enters an assessment of what it contains (e.g. if it is gunfire, and how many rounds). If the analyst confirms the report, an alert is sent as a text message (I believe in an app they furnish) to staff in APD's dispatch center, called the Emergency Communications Center (ECC). The contract includes an SLA on this process of I think 1 or 2 minutes, but I was told that they routinely performed as well as 30 seconds. Some APD personnel, I think usually area commanders but it may have been all field division sergeants, also receive the alerts on their phones.
The ECC dispatches the call as a priority 2. P2 is high enough that a ShotSpotter report will "bump" most calls for service except for a caller on the line violent crime in progress. When the officers arrive at the reported location, they make a brief assessment and search the area for suspects or victims. If no suspects or victims are found, a Crime Scene Technician is dispatched (often later as they will wait for daylight) to search the area for evidence such as spent shell casings.
My recollection is that I was told they were able to find definitive evidence of actual shots fired in less than five percent of cases, but take that with a grain of salt as I do not believe it was ever put in writing (I don't think they're allowed to by their contract) and I could be remembering wrong. However, it's believable that the accuracy of the system is higher than that suggests, as Albuquerque has a lot of wide open spaces that are difficult to search thoroughly if the ShotSpotter location estimate is at all inaccurate.
I was told that, when the system was tested by firing blanks, it was not completely effective but that they were satisfied with how effective it was. I was never given a number and I think they had been very specifically prohibited from discussing the testing in detail when they coordinated it with SoundThinking.
One of the major criticisms of the ShotSpotter system in Albuquerque is that it results in a relatively large volume of P2 calls that delay police response to most other calls for service. During the worst of the understaffing, I was told that some officers in high-crime areas like the International District spent a large portion of their total shift following up on ShotSpotter activations while there were multiple P3 calls queued. This has probably improved as staffing levels have increased, but in my mind it is the greatest single concern about the system.
SoundThinking's evasiveness, refusal of independent research, and clear motive to sell their product creates an alarming possibility that they are deceiving police leadership and elected officials into overprioritizing ShotSpotter. It may be a waste of money, which is already a problem, but the much greater concern is that police departments are putting off responding to nonviolent crimes in progress to go to ShotSpotter reports instead, because they have been told by SoundThinking that the accuracy rate of the system is very high.