BCSS Provides Fundamental Marine Logistics Support in Pioneering Deep Learning Framework for Marine Mammal and Bioacoustic Research – New Publication Now Out

BCSS is pleased to have collaborated with the Deep Voice Foundation (N. Bressler, M. Faran, A. Galor, M. M. Michelashvili, T. Nachshon, and N. Weiss) and facilitated a data acquisition expedition for the development of Soundbay, an open-source Python framework designed to facilitate bio-acoustic and machine learning research. Soundbay allows researchers to harness deep learning-based algorithms for acoustic audio analysis, providing an intuitive platform for the application of existing models and the effortless creation of new ones. The BCSS marine research logistics team at our Ocean Observatory played a vital role in providing research logistics support during the field work in Mozambique waters in 2018 for this project.

“The BCSS collaborated with the Deep Voice Foundation during several expeditions of acoustics data acquisition expedition in the Bazaruto archipelago in Mozambique in 2018. This region is a crucial breeding ground for the C1 sub population of humpback whales, which are known for their complex vocalizations.”

Photo: Deep Voice

Soundbay’s innovative features include the capability to compare baselines on different benchmarks, which is essential for emerging research and development in the field of deep-learning algorithms for animal call analysis. This framework’s versatility enables it to be applied to various domains and datasets, making it a valuable tool for bioacoustics research.

Read also: Spectacular Humpback Whale Behaviour Prior to Returning to their Feeding Grounds  

The BCSS collaborated with the Deep Voice Foundation during several expeditions of acoustics data acquisition in the Bazaruto archipelago in Mozambique in 2018. This region is a crucial breeding ground for the C1 sub population of humpback whales, which are known for their complex vocalizations. The recordings collected during this expedition encompass whale songs and social calls of humpback whales (in particular, mother and calf), as well as vocalizations from three species of dolphins. The data also include environmental noise such as coral reef noise, background noise from various sources including shipping, and equipment interactions with the surroundings.

Play Video
This video shows snippets of the expeditions of the Deep Voice foundation. This data does not seamlessly match the mentioned expeditions.
Play Video

This video shows snippets of the expeditions of the Deep Voice foundation. This data does not seamlessly match the mentioned expeditions.

Photo: Daphna Stern

The Deep Voice Foundation identified and labeled calls from the dataset using the Raven Lite 2 program, distinguishing between whale songs, whale social calls, dolphin calls, and other noises. These annotations serve as the positive class for the detection task, with other sounds classified as background noise. The data were meticulously divided into training, validation, and test sets to ensure no data contamination.

The collaboration between the BCSS and the Deep Voice Foundation resulted in the creation of an essential benchmark for marine mammal vocalizations detection. This benchmark, along with the Soundbay framework, holds promise for the marine mammal bioacoustics community, offering a tool that can drive the development of automated vocalization detection models, ultimately enhancing our understanding of marine mammals’ communication.

Read also: The Role of Humpback Whales in the Bazaruto Archipelago

Photo: Daphna Stern

Researchers are committed to the ongoing development and expansion of the Soundbay framework, with plans to include state-of-the-art architecture, an active-learning interface, and a broader range of bio-acoustic applications. The integration of Soundbay into an end-to-end service by Deep Voice will empower bioacousticians with the tools needed for efficient and reliable sound analysis, providing valuable insights for conservation and wildlife protection efforts.

For more information about the Soundbay please visit the framework via https://github.com/deep-voice/soundbay

The information provided in this release is based on the article “Soundbay: Deep Learning Framework for Marine Mammals and Bioacoustic Research” and its contents.

For questions about this article, please contact:
Ekaterina Kalashnikova, Bazaruto Center for Scientific Studies
Ekaterina.Kalashnikova@bcssmz.org

Bazaruto Center for Scientific Studies
Host of the first permanent Ocean Observatory focused on multi-ecosystem time series research in Africa, the Bazaruto Center for Scientific Studies (BCSS) was established in 2017 as in independent, non-profit organisation with a mission to protect and support the fragile ecosystems of the Bazaruto Archipelago, Mozambique. The research station is located on Benguerra Island, off the coast of Mozambique.

Follow us

Scroll to Top

Stay in the know

Receive monthly updates about our research, projects, new papers and spectacular wildlife encounters, straight into your inbox.