Quantifying bat call detection performance of humans and machines
Mark D Skowronski1, M Brock Fenton
1Department of Biology, University of Western Ontario, London, Ontario, Canada. mskowro2@uwo.ca
The Journal of the Acoustical Society of America
|January 29, 2009
Summary
Automated bat echolocation call detection methods offer faster processing but lower accuracy than human analysis. Optimal detectors balance speed and accuracy, crucial for effective bat recording system design.
Area of Science:
- Bioacoustics
- Wildlife monitoring
- Acoustic signal processing
Background:
- Detecting bat echolocation calls in field recordings is vital for ecological studies.
- Current detection methods vary in performance, impacting the effective range of recording systems.
Purpose of the Study:
- To compare the accuracy and speed of different echolocation call detection methods.
- To quantify the trade-off between detector speed and accuracy.
- To inform the design of bat field recording systems.
Main Methods:
- Experiments used synthetic echolocation calls from five bat species.
- Evaluated human detection accuracy against automated methods: model-based, energy-based, and optimal linear detectors.
- Measured detection accuracy, processing speed, and effective recording range.
Main Results:
- Human accuracy was 89.7%, model-based 76.3%, energy-based 72.2%, and optimal linear detector 98.4%.
- Energy-based detectors were significantly faster than model-based and human analysis.
- Call bandwidth influenced detection accuracy, with narrowband calls being easier to detect.
Conclusions:
- Automated detectors offer speed advantages but generally lower accuracy than human analysis.
- The optimal linear detector demonstrated superior performance in both accuracy and efficiency.
- Findings provide critical data for optimizing bat detection systems and field study designs.
