Convention Express Paper 122

Audio Engineering Society

Presented at the 155

Convention

2023 October 25-27, New York, USA

This Express Paper was selected on the basis of a submitted synopsis that has been peer-reviewed by at least two qualified anonymous

reviewers. The complete manuscript was not peer reviewed. This Express Paper has been reproduced from the author’s advance

manuscript without editing, corrections or consideration by the Review Board. The AES takes no responsibility for the contents.

portion thereof, is not permitted without direct permission from the Journal of the Audio Engineering Society.

Close and Distant Gunshot Recordings for Audio Forensic

Analysis

Robert C. Maher

Electrical & Computer Engineering, Montana State University, Bozeman, MT 59717-3780

Correspondence should be addressed to rmaher@montana.edu

ABSTRACT

We describe contemporary forensic interpretation of multiple concurrent gunshot audio recordings made in

acoustically complex surroundings. Criminal actions involving firearms are of ongoing concern to law

enforcement and the public. The U.S. FBI annually lists 166,000 criminal incidents involving firearms each year.

Meanwhile, over 80% of the large general-purpose law enforcement departments in the U.S. now use audio-

equipped body-worn cameras (BWCs), more than 135 communities use ShotSpotter

gunshot audio detection

systems, and tens of millions of private homes and businesses now have round-the-clock surveillance camera

systems—many of which also record audio. Thus, it is increasingly common for audio forensic examination of

gunshot incidents to include multiple concurrent audio recordings from a range of distances. A case study example

is presented.

1 Introduction

More and more criminal justice proceedings involve

audio forensic evidence in general, and gunshot audio

recordings in particular, due to (i) the large number of

law enforcement incidents each year in the United

States involving firearm use [1], (ii) the widespread

use of ShotSpotter

acoustic gunshot detection

systems [2], (iii) the millions of newly installed

residential security and doorbell camera systems with

audio capability [3], and (iv) the fact that most

general-purpose law enforcement agencies in the U.S.

are now utilizing body-worn cameras (BWCs) for all

routine work [4]. In fact, law enforcement

investigators now routinely canvass neighborhoods

and seek access to private surveillance recordings

following shootings and other criminal and civil

incidents involving gunshot sounds.

This work extends our previously reported research

sponsored by the U.S. National Institute of Justice in

nearfield gunshot acoustical analysis [5][6][7][8]. In

our current research, we focus upon audio forensic

gunshot audio cases in which concurrent recordings

from close and distant locations are available. For

example, a microphone within a few tens of meters of

the firearm, such as from a body-worn camera or a

nearby residential surveillance system, and a

concurrent recording from a few hundred meters

away obtained by ShotSpotter, a commercial gunshot

detection and localization system. With the known

“ground truth” from incident reports and the audio

recorded near the incident scene, we compare and

correlate information from ShotSpotter to develop

general recommendations and best practices for

improved audio forensic examination of gunshot

sounds recorded in complex acoustical environments.

n this paper, we briefly review the acoustical

properties of firearm sounds close to the gun and at

greater distances. Next, we present a case study

involving gunshots from an incident in which nearby

home surveillance cameras capture the gunshot

sounds, while simultaneously a ShotSpotter system

also detects the gunshots. Finally, we offer several

Maher Close and distant gunshot recordings for audio forensic analysis

AES 155

Convention, New York, USA

October 25-27, 2023

Page 2 of 8

remarks and suggestions regarding this type of

analysis, which we feel will become increasingly

common in audio forensics for the reasons mentioned

above.

2 Gunshot sounds

A conventional firearm uses a cartridge containing

gunpowder affixed with a bullet. Triggering the

primer in the cartridge rapidly combusts the

gunpowder within the confining cartridge, and the hot

combustion gases expand rapidly behind the bullet,

abruptly forcing the bullet and a jet of gas out of the

muzzle. The resulting muzzle blast causes an acoustic

pressure wave forming a loud pop sound lasting only

a few milliseconds. The muzzle blast is notably

directional: the on-axis sound level is more intense

than the level off to the side or toward the rear of the

firearm.

n summary, gunshot muzzle blast sounds are (a) high

amplitude (140+ dB SPL); (b) short in duration (2-3

milliseconds); (c) directional (sound level varies as a

function of azimuth with respect to gun barrel); and

(d)

ecorded with echoes and reverberation (acoustic

“impulse response” of the physical surroundings).

e firearm itself may produce relatively subtle

mechanical sounds before, during, and after the bullet

is discharged. These sounds may include the

mechanical action of the gun, such as the trigger and

cocking mechanism, ejection of the spent cartridge,

and positioning of new ammunition. These

characteristic sounds may be of interest for forensic

study if the microphone is located sufficiently close

to the firearm to pick up the tell-tale sonic

information.

f the bullet emerges from the barrel traveling faster

than the speed of sound, the supersonic projectile

creates a ballistic shock wave. The shock wave itself

forms a radially propagating cone that trails the bullet

as it travels down range. The expanding face of the

shock wave cone moves through the air at the speed

of sound, and the passage of the shock wave may be

picked up by microphones located near the bullet’s

path [9].

2.1 Reflections and reverberation

Forensic audio recordings of gunshots generally

contain significant evidence of acoustical reflections

and reverberation. Distant recordings may not be line-

of-sight to the shooting location, so the effects of

diffraction and multi-path interference is also

expected.

As noted above, the muzzle blast sound produced by

a firearm lasts only a few milliseconds, but the

recorded gunshot sound often has energy lasting

hundreds of milliseconds due to the reverberation of

the recording scene. The acoustic clutter of the

reverberation can sometimes provide good

information about the acoustical surroundings, but, in

general, the specific acoustical characteristics of the

firearm itself are lost in the overlapping sound

reflections.

2.2 Effects of audio coding

Many sources of user generated recordings (UGR)

come from devices that use perceptual audio coding

(e.g., MP3 or MP4) between the microphone and the

digital storage system. Perceptual coders are designed

to maintain the perceived audio quality of the original

signal, but such coders are not intended to retain

objective waveform information that would be

desirable for forensic purposes, especially for

impulsive signals such as gunshots. Nevertheless,

perceptually-coded audio can provide useful timing

information within the constraints of the block size

and timing of the coding algorithm [10].

urrently, commercial gunshot detection systems

such as ShotSpotter appear to use conventional

uncompressed pulse-code modulation (PCM) to store

recorded audio. These recordings should therefore

have fewer concerns about waveform interpretation

and timing.

3 ShotSpotter system

The ShotSpotter system from SoundThinking

™

[2] is

a proprietary commercial system intended to detect

and localize the sound of a gunshot in the jurisdiction

of a law enforcement agency that subscribes to the

ShotSpotter service. The system consists of numerous

microphones and computer processing systems

(sensor nodes) installed and maintained by

ShotSpotter on rooftops, poles, and other structures in

the area of coverage. ShotSpotter’s dispatching

service, which uses information derived from the

acoustic sensors to make a judgement about the

occurrence of a sound that might be a gunshot,

provides an estimate of the geographic location of the

sound source. The proprietary system uses signal

processing algorithms and human listeners.

he geometric coordinates of the sensor nodes are

determined by ShotSpotter via an integrated GPS

receiver at each node. The local clock at each node is

also time synchronized via the integrated GPS

receiver. The sensor nodes include a processor system

Maher Close and distant gunshot recordings for audio forensic analysis

AES 155

Convention, New York, USA

October 25-27, 2023

Page 3 of 8

to record and analyze audio digitized from the

acoustic sensor system, and a communication system

links each sensor to ShotSpotter’s central office.

The basic theoretical principle of the ShotSpotter

system is that a loud, impulsive sound (such as a

gunshot) will travel through the air in all directions at

the speed of sound and will then be detected as the

sound arrives successively at the acoustic sensor

nodes in the vicinity. The time-of-arrival of the

impulsive sound at the various sensors depends upon

the distance between the sound source and the sensor,

the speed of sound in the vicinity, the presence of

structures, terrain, and other obstacles, and various

atmospheric effects such as wind and temperature

gradients.

The audio processing system at each sensor node

includes computer firmware that uses various

proprietary algorithms to estimate the likelihood that

a received sound is a gunshot, and not some other

common sound like a door slamming or a firecracker.

If the algorithm makes the determination that a

gunshot sound was observed, the system processes

the recorded waveform and estimates the arrival time

of the impulsive sound based upon the details of the

microphone signal, and this information is sent to the

ShotSpotter central office.

When the ShotSpotter central office has received

several contemporaneous reports from the acoustic

sensor nodes in a particular area, the central office

system automatically uses the time reports and the

known sensor locations to estimate the time

difference between the sound’s earliest arrival at a

sensor—presumably the sensor closest to the sound

source—and the arrival times at the other sensors.

The known location of each sensor and the measured

time-difference of arrival at each sensor can be used

to compute an estimate of the sound source location.

This mathematical computation is known as

multilateration [11].

In order to notify the law enforcement agency about

the shot detection so that the agency’s officers can be

dispatched, ShotSpotter reports an estimated location

in two dimensions (e.g. latitude and longitude). The

sound arrival times from at least three sensor nodes

are needed for the mathematical multilateration to

compute a set of possible sound source locations that

are consistent with the observed time-differences of

arrival, and at least four sensor nodes are needed to

reduce the ambiguity of the possible source positions

[12].

4 Case study: concurrent close and

distant gunshot recordings

An example gunshot audio recording scenario comes

from a case involving a ShotSpotter gunfire incident

activation and several home surveillance recordings.

The vicinity of the incident is shown in Figure 1.

The shooting incident appears to have started in the

rear of a residential dwelling where a party crowd had

gathered late in the evening. One of the surveillance

recordings (Camera 3) included audio of the gunshot

sounds, while two other surveillance recordings

(Cameras 1 & 2, video only) captured images of

firearms being discharged.

Based upon the available surveillance camera audio

recording (Camera 3), a total of 17 gunshots are

audible. The initial two gunshots, separated by about

5 seconds, occur clearly in the surveillance audio

recording.

The location where the shooting incident took place

was covered by a ShotSpotter gunshot detection

system. In fact, the sequence of shots resulted in three

separate ShotSpotter detection reports, as described

below. Thus, the incident includes both close and

distant recordings of the gunfire sounds.

In this incident, the entire sequence of gunshot sounds

was captured by the audio recording from Camera 3,

located approximately 50 meters north of the area of

the party. The audio recording from Camera 3 is

shown in Figure 2.

Maher Close and distant gunshot recordings for audio forensic analysis

AES 155

Convention, New York, USA

October 25-27, 2023

Page 4 of 8

Figure 1: View of the incident scene showing the general location of the surveillance cameras and the path of the

shooters moving through the area.

Figure 2: Audio recording and spectrogram from Camera 3 position. Total duration 33.5 seconds. Individual

gunshot reports 1-17 indicated manually. The three ShotSpotter activation detections are shown (SS1, SS2, SS3).

Maher Close and distant gunshot recordings for audio forensic analysis

AES 155

Convention, New York, USA

October 25-27, 2023

Page 5 of 8

By comparing the timing of the shots present in the

Camera 3 recording to the incident reports from the

ShotSpotter system, we can identify that the

ShotSpotter system did not activate on the first two

audible gunshots, and also missed shots 9, 10, 11, 13,

and 14, as seen in Figure 2.

The Camera 3 position recorded the gunshot audio,

but the accompanying video did not contain images

of the shooting scene. The Camera 1 and Camera 2

locations did capture images with visible muzzle

flashes and other evidence of the gunfire (e.g., Figure

3, Figure 4, and Figure 5), but Camera 1 and 2 did not

have audio recording capability.

Figure 3: Frame from Camera 1, handgun muzzle

flash (Shot 7 from Figure 2).

Figure 4: Frame from Camera 2, handgun muzzle

flash (Shot 9 from Figure 2).

Based upon the Camera 3 audio recording, the shots

visible in the Camera 1 and 2 videos, and other

evidence reported from the incident scene, we can

deduce the position of the firearms making all 17

audible shots. The shot locations are depicted in

Figure 6.

Figure 5: Frame from Camera 1, handgun muzzle

flash (Shot 10 from Figure 2).

Figure 6: Approximate location of shots based upon

video information (Camera 1 and 2) and related

evidence.

The ShotSpotter system issued three real time

incident reports related to this case. The incident

report SS1 detected six shots, which correspond to

shots 3-8 in the sequence. Incident report SS2

indicated one gunshot, corresponding to shot 12, and

report SS3 indicated three shots, corresponding to

shots 15-17. The locations identified by ShotSpotter’s

multilateration for the ten detected gunshots are

shown in Figure 7.

In this case, it is clear that ShotSpotter did not detect

all of the shots, and of the shots it did detect, the

locations differ somewhat from the “ground truth”

provided by the close cameras. While this may appear

to call into question the value of ShotSpotter for

comprehensive audio forensic assessment of a

shooting incident, it is important to note that the real

time detections and locations sent by ShotSpotter to

Maher Close and distant gunshot recordings for audio forensic analysis

AES 155

Convention, New York, USA

October 25-27, 2023

Page 6 of 8

the law enforcement agency happened within one

minute after the incident, and the position information

was adequate to dispatch officers to the correct

residential block. So, while the accuracy of the

ShotSpotter multilateration positions in this example

show a noticeable discrepancy for crime scene

reconstruction purposes, the system did fulfill the

original intent of ShotSpotter, namely, to identify the

incident location very quickly and with sufficient

precision to dispatch the law enforcement responders

to the scene.

Figure 7: Location of shots from ShotSpotter

detection reports. The ShotSpotter positions are

within approximately 20 meters of the actual

gunshot locations observed via Camera 1 and 2.

Figure 8 shows an example excerpt from ShotSpotter

report SS1, identifying shots 3-8 of the incident. The

report includes excerpts of the audio recorded at just

three sensors, although it is assumed that ShotSpotter

had additional sensors used in the multilateration

calculation that are not listed in the summary report.

For the three sensors included in the summary report,

the table indicates the distance from each sensor to

the shot location determined by the multilateration

algorithm. ShotSpotter generally does not identify the

precise location of its sensors due to concerns about

vandalism. In any case, it is evident that the gunshot

acoustical waveforms differ among the three sensors

due to the different range, azimuth, elevation,

obstructions, reflections, and other physical path

differences.

Figure 8: Example excerpt from ShotSpotter

incident report SS1 (shots 3-8).

In this example, the three sensors in the example were

at distances 195, 565, and 856 meters, respectively

(see Figure 9). The ShotSpotter recordings are

significantly farther away from the firearms than the

roughly 50 meters for the Camera 3 surveillance

recording, and therefore the distant recordings will

include the effects of acoustical

diffraction/shadowing and reflections/reverberation

due to the numerous buildings, trees, and other

obstacles between the shooting location and the

ShotSpotter sensor microphones.

Figure 9: Aerial view of the neighborhood

surrounding the shooting incident, with distances to

the three ShotSpotter sensors indicated in summary

report SS1 (Figure 8).

Comparing the audio recordings from the close

microphone (Camera 3) and the more distant

ShotSpotter sensors, the differences in amplitude and

reverberation characteristics become apparent, as

presented in Figure 10.

Maher Close and distant gunshot recordings for audio forensic analysis

AES 155

Convention, New York, USA

October 25-27, 2023

Page 7 of 8

Figure 10: Comparison of the waveforms for shots

3-8 recorded by the close Camera 3 microphone and

the three more distant ShotSpotter sensors.

5 Discussion and Conclusions

This case study presents an increasingly common

situation in which a shooting incident is recorded by

a microphone close to the scene and concurrently by

microphones located some greater distance away. The

“close” recording in this case was from a home

surveillance system with audio recording capability,

while the distant recordings were obtained by the

commercial ShotSpotter gunshot detection system

deployed in the particular jurisdiction.

The close recording is useful for distinguishing

gunshots from various other sounds at the incident

scene, and for establishing a reliable timeline.

The close recording is susceptible to amplitude

clipping and other waveshape distortions due to the

sound intensity of firearm muzzle blasts, and the

likelihood that the recording was captured with

perceptual audio coding (e.g., MP3, MP4, etc.).

The more distant recordings, such as the ShotSpotter

recordings from 195, 565, and 856 meters available

in this case, have the advantage that amplitude

clipping is unlikely. The ShotSpotter system also

provides multilateration calculations that can help

estimate the location of the gunfire with respect to the

precisely known positions of the sensors, along with

precisely known timestamps. However, as was

observed in this case, ShotSpotter’s algorithms may

not correctly identify every gunshot sound in a

shooting incident, and the multilateration position

uncertainty may be greater than the dimensions of the

shooting scene, particularly if the shooters are in

obstructed locations or are moving during the

incident. Based on this case study, the audio forensic

recommendation would be to compare and confirm

the ShotSpotter information with the close recording,

but clearly the timing and details of each shot in the

incident is represented fully with the close recording,

and incompletely with the ShotSpotter reports.

While this example includes ShotSpotter evidence, it

is also increasingly common to have user-generated

surveillance recordings or cellphone video from

locations near and far from a shooting incident, or the

close recording may come from a law enforcement

body-worn camera system. The additional

information provided by the multiple concurrent

recordings may be important in answering questions

about the circumstances of the incident, but the fact

that the locations of user devices is usually not known

very precisely, it is generally not feasible to get

reliable multilateration estimates of the gunshot

location using muzzle blast time-of-arrival in the

various recordings. Nevertheless, there may be useful

information available by comparing the timing of

multiple shots in the incident recordings, such as

addressing the sequence of shots if multiple shooters

were present [13].

References

[1] Federal Bureau of Investigation, Crime Data

Explorer, URL:

https://cde.ucr.cjis.gov/LATEST/webapp/,

September 1, 2023.

[2] SoundThinking

™

, ShotSpotter gunshot

detection technology, URL:

https://www.soundthinking.com/law-

enforcement/gunshot-detection-technology/,

September 1, 2023.

[3] Straits Research, "Doorbell Camera Market,"

URL:

https://straitsresearch.com/report/doorbell-

camera-market, September 1, 2023.

Maher Close and distant gunshot recordings for audio forensic analysis

AES 155

Convention, New York, USA

October 25-27, 2023

Page 8 of 8

[4] U.S. National Institute of Justice, "Research

on Body-Worn Cameras and Law

Enforcement," URL:

https://nij.ojp.gov/topics/articles/research-

body-worn-cameras-and-law-enforcement,

January 7, 2022.

[5] R.C. Maher and T.K. Routh, "Advancing

forensic analysis of gunshot acoustics,"

Preprint 9471, Proc. 139th Audio Engineering

Society Convention, New York, NY (2015).

[6] R.C. Maher and T.K. Routh, "Wideband audio

recordings of gunshots: waveforms and

repeatability," Preprint 9634, Proc. 141st

Audio Engineering Society Convention, Los

Angeles, CA (2016).

[7] R.C. Maher, "Shot-to-shot variation in

gunshot acoustics experiments," elib 20461,

Proc. 2019 Audio Engineering Society

International Conference on Audio Forensics,

Porto, Portugal (2019).

[8] B.F. Miller, F.A. Robertson, and R.C. Maher,

"Forensic handling of user generated audio

recordings," Preprint 10515, Proc. 151st

Audio Engineering Society Convention,

Online (2021).

[9] R.C. Maher and S.R. Shaw, "Deciphering

gunshot recordings," Proc. Audio Engineering

Society 33rd Conference, Audio Forensics—

Theory and Practice, Denver, CO (2008).

[10] R.C. Maher and S.R. Shaw, "Gunshot

recordings from digital voice recorders," Proc.

Audio Engineering Society 54th Conference,

Audio Forensics—Techniques, Technologies,

and Practice, London, UK (2014).

[11] R.C. Maher and E.R. Hoerr, "Audio Forensic

Gunshot Analysis and Multilateration,"

Preprint 10100, Proc. 145th Audio

Engineering Society Convention, New York,

NY (2018).

[12] R. Calhoun and S. Lamkin, "Determining the

Source Location of Gunshots From Digital

Recordings," Express Paper 18, Proc. 153

Audio Engineering Society Convention, New

York, NY (2022).

[13] R.C. Maher, "Forensic interpretation and

processing of user generated audio

recordings," Preprint 10419, Proc. 149th

Audio Engineering Society Convention, New

York, NY, Online (2020).