More than a whistle: Automated detection of marine sound sources with a convolutional neural network

Abstract

The effective analysis of Passive Acoustic Monitoring (PAM) data has the potential to determine spatial and temporal variations in ecosystem health and species presence if automated detection and classification algorithms are capable of discrimination between marine species and the presence of anthropogenic and environmental noise. Extracting more than a single sound source or call type will enrich our understanding of the interaction between biological, anthropogenic and geophonic soundscape components in the marine environment. Advances in extracting ecologically valuable cues from the marine environment, embedded within the soundscape, are limited by the time required for manual analyses and the accuracy of existing algorithms when applied to large PAM datasets. In this work, a deep learning model is trained for multi-class marine sound source detection using cloud computing to explore its utility for extracting sound sources for use in marine mammal conservation and ecosystem monitoring. A training set is developed comprising existing datasets amalgamated across geographic, temporal and spatial scales, collected across a range of acoustic platforms. Transfer learning is used to fine-tune an open-source state-of-the-art ‘small-scale’ convolutional neural network (CNN) to detect odontocete tonal and broadband call types and vessel noise (from 0 to 48 kHz). The developed CNN architecture uses a custom image input to exploit the differences in temporal and frequency characteristics between each sound source. Each sound source is identified with high accuracy across various test conditions, including variable signal-to-noise-ratio. We evaluate the effect of ambient noise on detector performance, outlining the importance of understanding the variability of the regional soundscape for which it will be deployed. Our work provides a computationally low-cost, efficient framework for mining big marine acoustic data, for information on temporal scales relevant to the management of marine protected areas and the conservation of vulnerable species.

Major Soundscape Differences in the West Coast of Scotland Are Reflected in Acoustic Indices

“…A whistle label could include all other labels, click labels excluded whistles, vessel labels excluded clicks and whistles, and ambient labels excluded the other three. The labelling processes was designed for a CNN model 23 , with 3 second labelling windows chosen as a compromise between the time periods of milliseconds for clicks, seconds for whistles and minutes for a vessel passage. Data were labelled both visually using spectrograms, and aurally using Audacity Software (Audacity version 3.0.02, 2021).…”

“…Data was labelled between the hours of 18:00 and 09:00 for each day to maximise the detection of delphinid activity. Mean error rate in the labelling was estimated to be 3.3%, with the greatest error occurring between Ambient (6.3%) and Vessel (3.7%) classes (see 23 for full details).…”

Bioacoustic Event Detection with Self-Supervised Contrastive Learning

“…Large-scale studies often employ Convolutional Neural Network (CNN) architectures to carry out supervised detection or classification tasks across diverse species. These include detection of humpback whale vocalizations in passive acoustic monitoring datasets 5 ; detection, classification, and censusing of blue whale sounds 6 ; avian species monitoring based on a CNN trained to predict the species class label given input audio 7 ; presence indication of Hainanese gibbons using a ResNet-based CNN 8 ; and, recently, detection and classification of marine sound sources using an image-based approach to spectrogram classification 9 . However, such supervised learning approaches remain limited in their scope, hindering their capacity to be deployed in real-time data processing pipelines.…”

Who Goes There? Evaluating the Performance of a CNN for Monitoring Delphinid Presences to That of the C-Pod Within Diverse Marine Soundscapes

“…Automation of the detection of these acoustic signals is therefore an essential component of the analysis pipeline for gathering data on delphinid occurrence.The application of Deep Learning (DL) algorithms is increasingly prevalent within the field of bioacoustics. Convolutional Neural Networks (CNNs), which use image processing techniques to read and classify acoustic data into predetermined classes, have been shown to excel at detecting signals which vary in respect to time, frequency, and amplitude 13,14,15,16 . Using broadband acoustic data as input, CNNs have achieved high accuracies at the task of detecting dolphin vocalisations within variable marine environments 17,18,19,20 .…”

“…Proprietary software is provided which contains a custom classifier (KERNO) to classify click trains as originating from either dolphins, porpoises or other cetaceans based on their intensity, duration, frequency content and inter-click intervals (ICI) 31,32 . In this work we evaluate the performance of the C-POD compared to a newly available multi-sound source CNN 16 for the task of monitoring hourly dolphin presence in the waters off the west coast of Scotland (Figure 1), using PAM data collected within the COMPASS array (EU INTERREG COMPASS project). We present the first empirical comparison between a CNN and the C-POD for dolphin detection, highlighting the efficiency of each method as a monitoring tool in diverse soundscape conditions.…”