Summary: A new University of Exeter study shows African lions produce two distinct roar types — the well-known full-throated roar and a newly identified intermediary roar. Using machine learning and a data-driven approach, researchers automatically classified these vocalizations with high accuracy, reducing human bias and improving passive acoustic monitoring for conservation.
This breakthrough strengthens noninvasive methods to estimate population size and track individuals, offering conservationists a more reliable tool as lion numbers decline across Africa. By combining bioacoustics with AI, the study demonstrates how automated sound analysis can support targeted protection strategies for vulnerable big-cat populations.
Key facts
- Two roar types identified: Lions produce both full-throated roars and a distinct intermediary roar within a single roaring bout.
- High AI accuracy: Simple acoustic metrics combined with K-means clustering classified roar types with 95.4% accuracy; a two-state Hidden-Markov Model also detected two roar states with 84.7% accuracy.
- Improved individual identification: Data-driven selection of full-throated roars improved individual recognition (F1-score 0.87) compared with manual classification (F1-score 0.80).
- Conservation implications: Automated acoustic monitoring offers scalable, less biased population estimates and complements camera traps and field surveys.
Overview of the study
Researchers from the University of Exeter led a collaborative project involving multiple conservation and research organizations to reassess what constitutes a lion’s “roar.” Historically, studies and surveys focused on a single iconic full-throated roar. The new research shows that a roaring bout typically contains two acoustically distinct elements: the full-throated roar and an intermediary roar. Recognizing both types improves the accuracy and consistency of acoustic monitoring.
Methods and results
The team applied machine learning techniques to recorded roaring bouts, using straightforward acoustic features — maximum frequency (Hz) and call duration (s) — together with K-means clustering to separate roar types. This simple, interpretable approach yielded 95.4% classification accuracy. Complementary modelling with two-state Gaussian Hidden-Markov Models supported the existence of two distinct call types within bouts and produced an 84.7% classification rate for the two roar states.
Automating roar classification reduces reliance on expert judgment, thereby lowering human-induced bias in selecting full-throated roars for individual identification. When the research team used model-predicted full-throated roars, the system identified individuals more reliably than when manual selection was used.
Why this matters for conservation
The International Union for Conservation of Nature lists African lions as vulnerable, with wild populations estimated at 20,000–25,000 animals and a marked decline over recent decades. Passive acoustic monitoring equipped with validated AI classifiers offers a scalable, noninvasive way to detect, count, and monitor lions across large and remote landscapes. This method can supplement camera surveys, spoor tracking, and other traditional techniques, helping conservationists allocate resources and measure the effectiveness of protection strategies.
Expert perspective
Lead author Jonathan Growcott explained that lion roars carry unique signatures useful for estimating population size and tracking individuals. By removing much of the subjectivity in call selection, automated bioacoustic pipelines can deliver more consistent data for ecological research and conservation planning.
Collaborators and funding
The research was carried out in collaboration with the Wildlife Conservation Unit at the University of Oxford, Lion Landscapes, the Frankfurt Zoological Society, TAWIRI (Tanzania Wildlife Research Institute), TANAPA (Tanzania National Parks Authority), and computer scientists from Exeter and Oxford.
Funding: Supported by the Lion Recovery Fund, WWF Germany, the Darwin Initiative, and the UKRI AI Centre for Doctoral Training in Environmental Intelligence.
FAQ
- Q: What new discovery was made about lion vocalizations?
A: Researchers identified a second, distinct intermediary roar in addition to the classic full-throated roar. - Q: How does AI improve lion monitoring?
A: Machine learning classifies roar types automatically with high accuracy, reducing human bias and improving individual identification. - Q: How will this help conservation?
A: Precise acoustic tracking offers a scalable, noninvasive method to estimate population sizes and monitor trends, supporting targeted protection of declining lion populations.
Abstract (condensed)
African lions emit roaring bouts used for territorial advertisement and intra-pride communication. This study proposes a data-driven method to automatically classify full-throated roars within the broader roaring bout, revealing a second, intermediary roar type. Using simple acoustic metrics and clustering, the approach achieves 95.4% classification accuracy and improves individual identification performance. The research outlines a clear, implementable workflow to expand passive acoustic monitoring, making it a practical tool for large-scale lion conservation research.
Editorial notes
- Article edited for clarity and context by editorial staff.
- Journal paper reviewed in full by the reporting team.
Author: Louise Vennells
Source: University of Exeter
Contact: University of Exeter press office
Image credit: Neuroscience News