Workflow and convolutional neural network for automated identification of animal sounds

The use of passive acoustic monitoring in wildlife ecology has increased dramatically in recent years as researchers take advantage of improvements in autonomous recording units and analytical methods. These technologies have allowed researchers to collect large quantities of acoustic data which must then be processed to extract meaningful information, e.g. target species detections. A persistent issue in acoustic monitoring is the challenge of efficiently automating the detection of species of interest, and deep learning has emerged as a powerful approach to accomplish this task. Here we report on the development and application of a deep convolutional neural network for the automated detection of 14 forest-adapted birds and mammals by classifying spectrogram images generated from short audio clips. The neural network performed well for most species, with precision exceeding 90% and recall exceeding 50% at high score thresholds, indicating high power to detect these species when they were present and vocally active, combined with a low proportion of false positives. We describe a multi-step workflow that integrates this neural network to efficiently process large volumes of audio data with a combination of automated detection and human review. This workflow reduces the necessary human effort by > 99% compared to full manual review of the data. As an optional component of this workflow, we developed a graphical interface for the neural network that can be run through RStudio using the Shiny package, creating a portable and user-friendly way for field biologists and managers to efficiently process audio data and detect these target species close to the point of collection and with minimal delays using consumer-grade computers.