Machine Learning Breakthrough Allows Lion Roar Tracking via Acceleration Sensors Without Using Microphones

Scientists use machine learning to identify lion roars from motion sensors, enabling long-term pride tracking without using microphones or audio files.

By: AXL Media

Published: Apr 23, 2026, 6:08 AM EDT

Source: Information for this report was sourced from EurekAlert!

Decoding Infrasonic Patterns Through Motion Sensors

Researchers at the Leibniz Institute for Zoo and Wildlife Research have introduced a transformative method for monitoring lion vocalizations that bypasses the limitations of traditional acoustic recording. By utilizing a specialized machine learning algorithm, the team can now detect long-distance roaring based exclusively on acceleration data (ACC) captured by wearable collars. Unlike microphones, which consume significant power and storage, ACC sensors record three-dimensional movement patterns over extended periods. Dr. Ortwin Aschenborn, a wildlife veterinarian with the GAIA Initiative, noted that while the acoustic properties of roars are well-documented, the spatial and social functions of these sounds, particularly among female lions, remain largely unexplored due to previous technological constraints.

Superior Accuracy Across Gender and Movement States

The newly developed algorithm, known as a "fully convolutional neural network" or U-Net, represents a significant upgrade over previous models that only functioned when animals were stationary. This AI can successfully isolate the subtle, fine-scale vibrations of a roar even when mixed with the rhythmic patterns of a lion walking. Wanja Rast, an AI specialist at Leibniz-IZW, explained that training the model involved synchronizing 1,333 recorded roaring events from lions in Etosha National Park. The resulting system achieves an accuracy rate between 90 and 96 percent, effectively distinguishing genuine roars from other physical activities for both male and female lions.

Optimizing Field Research Through Low Energy Data

The shift from audio loggers to acceleration sensors addresses a major logistical hurdle in wildlife biology: the high energy demand of recording sound. Acceleration data requires far less storage space, allowing researchers to track pride members over several months without frequent battery replacements or data retrieval missions. Dr. Jörg Melzheimer highlighted that this method not only supports new studies but can also be applied retrospectively to existing datasets that were not originally intended for vocalization research. While not every species produces a physical vibration strong enough to be detected via ACC sensors, the success with lions provides a foundational framework for similar applications in other vocal mammals....