The Encore Nobody Was Listening For
African lions do not stop at one roar. A study of 3,149 recorded vocalizations found that every full-throated roar is followed, without fail, by a second, quieter call, a pattern that went unnoticed for as long as anyone has been recording lions. The team, led by Jonathan Growcott, a doctoral student in mathematics and statistics at the University of Exeter, and colleagues from the University of Oxford, Lion Landscapes, Frankfurt Zoological Society, TAWIRI and TANAPA, published the finding on Nov. 20, 2025, in the journal Ecology and Evolution, in a paper titled “Roar Data: Redefining a Lion’s Roar Using Machine Learning.”
The sound everyone already knows is the full-throated roar, the one that opens every nature documentary: loud, complex, arcing up and down in pitch. What comes after it, Growcott says, is flatter and lower, “a flatter sound with less variation.” He and his coauthors named it the intermediate roar, and it turns up after the main event again and again, across two countries.
One video filmed in Zimbabwe’s Bubye Valley Conservancy, credited to Matt Wijers, catches the full sequence in a single lioness: low, rising moans building into a full-throated roar, a pause, then this newly named intermediate roar, and finally a string of closing grunts. That is one animal, one clip, but it is also exactly the pattern the 3,149-call dataset shows over and over.
How the AI Caught What Human Ears Missed
The recordings came from two very different setups. In Tanzania’s Nyerere National Park, the team scattered 50 custom-built microphones across the park and let them run unattended. In Zimbabwe’s Bubye Valley Conservancy, five lions wore acoustic sensors built into their collars, so the microphone traveled with the roar. Together, the two sites produced the 3,149 vocalizations behind the study.
Sorting that much sound by ear is where the old approach broke down. A roaring bout moves fast, and the intermediate call is easy to mistake for the tail end of the main roar rather than a thing of its own. The researchers ran the recordings through a two-state Gaussian Hidden-Markov Model, a statistical tool that tracks how a sound shifts from one acoustic state to another, and asked it to separate full-throated from intermediate roars within each bout. It managed the split at 84.7% accuracy. A simpler approach, using just two measurements per call, maximum frequency and call length, sorted lion calls into their types with 95.4% accuracy through a clustering method.
None of it ran on its own. Researchers still had to mark where each roar started and stopped and strip out the moans that typically lead into a roaring bout before the model could learn the difference. The AI did not replace human ears here; it gave them something sharper to listen with.
Why Roar Twice
Earlier research had already shown that a lion’s full-throated roar works like an ID card thrown into the dark: it carries information about an animal’s sex, age and other individual traits, on top of reaching pride members and marking territory. Lions make plenty of other sounds too, grunts, snarls, chuffs, moans, mews, but the full-throated roar is the only one shown to hold that kind of identifying signature.
What the intermediate roar is for is not established. What stands out is how reliably it shows up: after full-throated roars, in Tanzania and in Zimbabwe, in a free-ranging population and in collared individuals, not as a random fade-out but as a repeatable second step. That consistency is itself a finding, even with no confirmed function attached to it yet, and it argues against the intermediate roar being acoustic leftover.
Separating the two roar types cleanly also sharpened a practical measurement. Using the AI’s predicted full-throated roars, stripped of the intermediate call that used to blur into them, improved the researchers’ ability to identify individual lions from their calls, with an F1 score, a combined measure of how often identifications are both correct and complete, of 0.87 against 0.80 for the older, manually classified roars.
The One-Roar Lion, Retired
The long-standing assumption, repeated in documentary narration and casual conversation alike, was that a lion has one kind of roar. It does not hold up. The study identifies two structurally different roar types inside a single roaring bout, the full-throated roar and the newly named intermediate roar, occurring in a reliable sequence rather than as variations on one sound. The one-roar model is not a simplification anymore. It is just wrong.
What the Roar Still Won’t Say
Whether the intermediate roar carries its own identifying signature, the way the full-throated roar does, is not yet known. Neither is its specific behavioral job. Asked what it is for, Growcott told Science News: “That part is still a mystery. We don’t know yet. Unfortunately, we don’t speak lion. There is no option of ‘lion’ on Duolingo.”
The pattern has so far only been documented in two populations, one in Tanzania and one in Zimbabwe, and whether it holds across lion populations elsewhere is an open question, along with whether individual sub-sections of a roaring bout carry their own specific information.
There is a reason beyond curiosity to keep testing this. With roughly 23,000 African lions left in the wild, per the IUCN Red List, conservationists lean on methods like camera traps and spoor surveys, both resource-heavy and not always precise, to estimate population sizes. Growcott’s hope, not yet a result, is that cleanly telling the two roar types apart could eventually make acoustic surveys sharper for counting lions than either method manages now.