--
Have you heard about what ELP’s been up to lately? ELP researchers¹ have just released a paper describing our automated system for detecting forest elephant calls, also referred to as “rumbles.” Since 2000, ELP has collected over 750,000 hours of forest sounds from Gabon, Congo-Brazzaville, Cameroon, and Central African Republic. Peter Wrege, ELP’s director says “To my knowledge, this is the largest collection of forest recordings from anywhere in the world and certainly the largest collection of forest elephant vocalizations.”
The data are collected by automatic recording units (ARUs) installed on trees located in the forests where elephants roam. Although deploying and maintaining these recorders in the forest can be a challenge, managing the actual sound files takes a tremendous amount of human effort and time. Until recently our technicians examined hours of recordings manually, a process so time-consuming that ELP could analyze only 10% of the data collected by the ARUs². ELP needed a new system to keep from drowning in data as the number of study sites increased. After many months and tests, researchers developed an algorithm for the automated detection of elephant rumbles in sound recordings. With the new automated detection system, ELP can review every hour on file, making the detection system at least 300 times more efficient than the old hand-browsing method. The tradeoff is a slightly lower accuracy rate because the detector does miss tagging some elephant rumbles. After the automated detector locates the distinct forest elephant vocalizations — low frequency, slightly curved tones with harmonics — in hours-worth of data, ELP workers and volunteers check the count for accuracy, because occasionally the detector will mistake thunder, vehicles, nearby insects, and even crocodiles for elephant rumbles. However, the ability to use all of the data to analyze elephant behavior makes this detector invaluable, and ELP is always working to improve the system. The detector is also uniquely capable of picking out elephant rumbles amidst the vast array of rainforest noises and conditions, and opens up the possibility of a radical change to wildlife monitoring beyond the elephant world. The authors hope that their detection algorithm will aid other conservation efforts.
References:
Keene et al. “Automated detection of low-frequency rumbles of forest elephants: A critical tool for their conservation.” April 18, 2017. The Journal of the Acoustical Society of America 141, 2715 (2017). doi: http://dx.doi.org/10.1121/1.4979476
