New AI Breakthrough Uses Single Blue Whale Call to Unlock 25 Years of Hidden Underwater Acoustic Data

Researchers at UNSW used a single blue whale call to train AI that detects songs across 25 years of ocean data with 99.4% accuracy, aiding rare species tracking.

By: AXL Media

Published: May 1, 2026, 11:35 AM EDT

Source: Information for this report was sourced from Earth.com

Revolutionizing Marine Bioacoustics with Minimal Data

The vast archives of the world’s oceans contain decades of continuous audio, much of which remains unanalyzed due to the sheer volume of data and the scarcity of known animal vocalizations. Traditionally, training artificial intelligence to recognize specific species required thousands of verified examples—a luxury rarely available for endangered animals. However, a study published in Nature by a team at UNSW has demonstrated that a single, high-quality recording of a blue whale song can be transformed into a scalable search tool. This breakthrough allows researchers to turn overlooked hydrophone recordings into vital evidence for species monitoring and conservation.

The Power of Data Augmentation and Transfer Learning

The technical success of the project, led by PhD candidate Ben Jancovich, relies on a process known as data augmentation. Instead of searching for new calls, the team took one clean recording and created numerous realistic variations by altering pitch, duration, and background noise levels. These synthetic examples taught the neural network to recognize the whale’s signature song even when distorted by distance or ocean interference. Furthermore, the team utilized transfer learning by adapting AI architectures originally designed for human speech recognition, allowing the model to be trained on a standard laptop in just a few hours.

Consistent Song Patterns of the Chagos Pygmy Blue Whale

The method is particularly effective for blue whales because their populations sing highly consistent, repetitive songs. In the case of the Chagos pygmy blue whale, found in the central Indian Ocean, these stable acoustic features provided a precise target for the AI. During testing, the model achieved a 99.4% recall score, successfully identifying nearly every call within the dataset. While ocean noise remains a constant threat to these endangered mammals, according to NOAA Fisheries, this high-precision listening tool offers a non-invasive way to track their movements and recovery over vast geographical distances.