Working together with researchers from France, MERIDIAN is helping compile a catalog of all known fish sounds and creating a benchmark dataset for marine bioacoustics detection and classification algorithms. Read on to find out more.
Modern deep neural networks owe their success at solving tasks such as image classification and voice recognition to the existence of large training and benchmark datasets with thousands and even millions of labelled data samples, e.g., ImageNet. When it comes to underwater acoustics, however, there is still a long way to go in terms of compiling datasets of sufficient size and quality, as well as properly describing the datasets and making them openly available. Inspired by the tremendous impact ImageNet has had on the field of machine learning, MERIDIAN is working together with researchers from IMT Atlantique in France to create a similar benchmark dataset for marine bioacoustics with the name CetaSound. Focussing initially on marine mammals, the project makes use of publicly available acoustic data samples such as those found in the Watkins Marine Mammal Sound Database. If you want to contribute to CetaSound or just learn more, don’t hesitate to contact us!