The Burrunan dolphin (Tursiops australis) is a genetically distinct species with an isolated population of approximately 300 individuals in Port Phillip Bay. Listed as critically endangered under the Victorian Flora and Fauna Guarantee Act, this species faces numerous challenges, including anthropogenic noise, toxicants, and vessel impacts. Traditional visual observation methods have restricted the study of their behaviours, but the Marine Mammal Foundation (MMF) is leveraging acoustic technologies to conduct passive acoustic monitoring (PAM). Between 2018 and 2022, MMF collected 7,080 hours of recordings from 10 sites, generating over 2,000 annotations within 83 annotated hours. These efforts aim to reveal how dolphins use their environment on a 24-hour basis, their acoustic repertoire, and the impacts of human activity on their population.
With only 1% of the dataset annotated by researchers, our expertise became indispensable. Through advanced machine learning models, we automated the detection of dolphin vocalisations within the vast dataset. By tailoring algorithms to the unique acoustic environment of Port Phillip Bay, we provided MMF with a comprehensive list of files containing dolphin calls and their respective counts. This allows MMF to prioritise high-value recordings and focus on interpreting ecological patterns, ultimately informing targeted conservation strategies.
THE PROBLEM
Manual analysis of vast acoustic datasets is unfeasible for a small NGO like MMF. While PAM devices continuously record underwater sounds, generating invaluable data and detecting dolphin vocalisations among diverse marine sounds are challenging. The myriad data classes require slightly different preprocessing to suit the model, adding complexity to the workflow. Additionally, low signal-to-noise ratios for specific call types make identifying dolphin calls difficult, mainly when competing biological noises like shrimp clicks overlap. Finally, MMF focuses on identifying the most informative files with the highest presence of dolphin calls rather than tagging every individual call, making manual analysis even less practical. Crucial behavioural insights and conservation strategies remain out of reach without efficient tools.
THE SOLUTION
Working closely with MMF, we customised a machine-learning algorithm to process large acoustic datasets efficiently. This detector identifies diverse vocalisations while addressing challenges like incomplete tagging and low signal-to-noise ratios. Researchers can focus their manual efforts on the most informative recordings by grading files based on dolphin call presence. For the first time, the project has revealed Burrunan dolphin movements and behaviours across a 24-hour cycle, providing insights previously unattainable through visual observations. This streamlined process enables MMF to continue collecting, annotating, and analysing new data, significantly advancing conservation management for this critically endangered species.
THE PROJECT’S TEAM
Shai Nahum Gefen
ML Researcher
Dr Kate Robb
Executive Director of the Marine Mammal Foundation
Amber Crittenden
Research Associate at the Marine Mammal Foundation
Our people are important to us much as the whales!
WHO WE ARE?
Our people are important to us much as the whales!
Our volunteers are motivated and collaborative and share a passion for solving problems and the sea. We seek individuals inspired by the prospect of developing new technology and driven by the will to make on environmental change. Deep Voice includes five teams: Research, Bio-annotations,Development, Product, and Media.
