Gunshot detector — tracking audio signals through the forests of Africa
Elephant Listening Project (ELP)’s acoustic recording units are being used to collect vast amounts of audio data from the forests of Gabon, Republic of Congo, Cameroon, and elsewhere with the goal of monitoring both elephant activity and poaching. After retrieving memory cards, we analyze all these data at Cornell. We run a gunshot search on the audio to help us measure poaching intensity in each recording area. These data help to inform anti-poaching efforts, for example, at Korup National Park where our data helped patrols prioritize high-poaching areas. I talked to postdoc Yu Shiu¹, who develops algorithms (systems) for detecting sounds of interest within the audio data. Here is what I learned…
It is easy for an algorithm to isolate sharp onsets (sudden sounds), but most of those, it turns out, are not actually gunshots. Yu Shiu is developing a better algorithm to specifically discern gunshots by examining the reverb immediately after the sharp onset. Gunshots generate a distinct lingering reverb around the 500 Hz range (low sounds) during this time. Using this, the algorithm can discern gunshots with reasonable accuracy. This will bring us closer to the goal of detecting poaching activity in real time, with high accuracy!
Yu Shiu is considering the potential of using other parts of the reverb for new functionality as well. The reverb could help us understand the terrain. The reason why you sound good singing in the shower is that your bathroom’s hard surfaces and tiles reflect sound and enhance reverberation². Singers hate singing in rooms with thick carpet because soft surfaces absorb sound and kill reverb. Similarly, a gunshot in a valley should be distinguishable by its enhanced reverb relative to a gunshot on a flat surface or at the top of a hill.
The theoretical next step would be to build the infrastructure to notify park officials as gunshots occur. The main stumbling blocks with this project are conserving battery life and real-time communication. In addition, false positives are much more harmful in a real-time environment because we risk “crying wolf” to authorities when there is no poaching. These unit prototypes are also much more expensive, so getting them to be as accurate and economical as possible is high on the priority list. ELP researchers are all working hard toward these goals.
¹April 5, 2017 Interview with Yu Shiu
²ISO 354:2003 Acoustics — Measurement of sound absorption in a reverberation room. Tables of absorption coefficients for various materials. The data reveals a general trend of soft surfaces absorbing greater sound.
Easy to follow explanation
Find out more about the Elephant Listening Project
Conserving the tropical forests of Africa through acoustic monitoring, sound science, and education, focusing on forest elephants