When a lion roars across the savannah, the sound can carry for miles, shaking air, grass, and bone alike. It's the ultimate symbol of dominance. But to the delight of scientists and data geeks alike, that earth-rattling call just got more complicated.
A team of ecologists and computer scientists from the University of Exeter and Oxford have discovered that the African lion doesn't have one roar -- it has two. By training an algorithm to analyze thousands of hours of audio from Tanzania and Zimbabwe, the researchers uncovered an "intermediary roar" that has been hiding in plain sound all along.
"Lion roars are not just iconic -- they are unique signatures that can be used to estimate population sizes and monitor individual animals," says lead author Jonathan Growcott from the University of Exeter.
"Until now, identifying these roars relied heavily on expert judgment, introducing potential human bias. Our new approach using AI promises more accurate and less subjective monitoring."
Growcott's team planted autonomous audio recorders, nicknamed CARACALs, throughout Tanzania's Nyerere National Park. For two months, these devices soaked up every nocturnal "podcast" the lions had, from low moans to grunts to the high drama of a full roar.
When the researchers ran the data through their model, they found something unexpected. Each "roaring bout" wasn't a single type of call -- it contained multiple, distinct components. There was the long-recognized full-throated roar, but right before the closing grunts came a shorter, lower-pitched sound that didn't match any known pattern.
In the study published in Ecology and Evolution, the team christened it the "intermediary roar." Using two simple sound measurements -- how long the roar lasts and how high its pitch climbs -- the model classified calls with 95.4% accuracy. That's better than human experts, whose manual labeling topped out near 84%.
The intermediary roar, says Growcott, "sits between the moans and the full-throated roars, like a bridge in the sonic architecture of the lion's call."
So, what does it mean? That part is still a mystery. "We don't know yet," Growcott told Science News. "Unfortunately, we don't speak lion. There is no option of 'lion' on Duolingo."
Yet that doesn't mean the findings can't be practical.
Lions have disappeared from more than 90% of their historic range. Fewer than 25,000 remain in the wild, their populations sliced apart by habitat loss, poaching, and human conflict. Counting them accurately is a logistical nightmare. Camera traps miss lions that roam at night, and footprint surveys aren't any better.
But roars carry for kilometers. And every lion's roar is as unique as a fingerprint. With AI listening posts scattered across the landscape, conservationists could track individuals without ever seeing them.
"If you can identify a lion by its roar, this could potentially be a tool to count the number of individuals within a landscape," Growcott told Science News.
To test that idea, the team compared AI-classified roars with those gathered from collared lions in Zimbabwe. The algorithm not only recognized the two roar types -- it could also tell lions apart, improving individual identification scores by 9% compared to manual classification.
The method is simple enough to use even in resource-limited field stations. It requires no massive neural networks or expensive GPUs, just clear audio and a few lines of code. "We believe there needs to be a paradigm shift in wildlife monitoring and a large-scale change to using passive acoustic techniques," Growcott says.
And, like human languages, lion roars even appear to vary by region. The study found differences in pitch and duration between lions in Tanzania and those in Zimbabwe. One individual from Botswana produced distinctly shorter, lower roars, perhaps a hint of local "lion accents."