How to Save the World’s Largest Lemurs? Listen to Their Songs.

The tropical rainforests of Madagascar are a raucous wash of sound, pierced by astonishingly loud, musical singing. These are the vocalizations of the indri, the largest species of lemur in the world. Emanating from up in the green-black canopy, the indri songs are notably mournful and resonant—partly whale song and partly wolf howl. They rise to a crescendo and then fall away into silence.

For roughly 88 million years, Madagascar has been geographically isolated from Africa and Asia. Considered something of an “evolutionary sandbox,” the ancestors of many of its terrestrial mammals are thought to have reached the island from continental Africa, some floating on mats of vegetation carried by ancient ocean currents. Once there, isolated populations proliferated and diversified, producing unique wildlife found nowhere else on Earth.

The indri is one such animal. Endemic to the island, the species is critically endangered; they’re threatened by habitat loss caused in part by tavy, the slash-and-burn agriculture that destroys their arboreal habitat. They’re especially vulnerable because all captive conservation attempts have failed, making the protection of wild indri the only viable path to their survival. Now, to estimate the population in the hope of conserving them, scientists are using a process called passive acoustic monitoring, or PAM, which measures indri density and abundance. 

“The indri is one of the most endangered primates on Earth,” says Valeria Ferrario, a postdoctoral fellow at the University of Turin and first author of a new study documenting PAM in indri populations. “Since this species doesn’t survive in captivity, the first step for saving it is to understand how many individuals remain in the wild. PAM gives us, for the first time, a scalable and noninvasive way to listen to what is left.”

In Maromizaha, a protected area in the eastern part of Madagascar, local Malagasy guides led researchers into the thicket of the forest at dawn, armed with conservation technology. The team—which included Filippo Carugati, a biologist from the University of Turin who co-authored the study—attached 26 autonomous recording units, or ARUs, to trees with plastic cords. ARUs are devices that automatically pick up forest sounds on a predetermined schedule. 

The devices recorded 10 minutes every half hour—a cadence meant to optimize battery life and data storage capacity. Researchers then used the captured songs to estimate the population of indri groups, which consist of two to four individuals, using an AI model trained to recognize their vocalizations. 

“Indris are the only lemurs that produce songs, using them to claim territory to other rival lemur groups,” says Carugati. “The songs are very complex vocal sequences with two individuals duetting with dimorphic notes between the male and the female.” 

When more than two indris in a group sing, it’s called a chorus. And if one group starts to sing, neighboring groups respond, and the vocalizations travel all over the forest like a cheer rolling through a stadium—a “chorus moving through the forest,” Carugati says. The team took note of several factors, such as where the chorus started and the direction of the singing. From there, researchers can estimate the groups of indris based on where they hear the sounds. 

However, there was a complication. Hearing one indri song isn’t the same as counting an indri group. The same family can sing more than once. Some songs, drowned out by the din of the forest, may not reach a microphone at all. To turn thousands of detections into an estimate of population density, the researchers needed to know both how frequently indri groups sing and how likely their microphones were to hear them. 

Ferrario solved this by using a visual representation of sound called a spectrogram to decode the vocalizations. A spectrogram is a digital graph that shows how the frequency and intensity of a sound evolves over time, letting researchers see which species are vocalizing, just by looking at the shape of the sound. 

On a spectrogram, the dark color of notes indicates the intensity—the nearer to the song being recorded, the darker the color. The further the song is, the notes are more of a light gray. “Based on the spectrogram, you can understand which group is singing,” says Ferrario. 

The team based its research on 12 families of indris whose identities and general locations were already known. When one group sang, the researchers knew the origination point and which ARUs had recorded it. They could then estimate the likelihood of picking up songs based on distance and orientation. 

The AI model identified 5,395 potential songs from the ARUs over 13 months, which researchers then corrected for false positives. Around 38 percent of detections were found to be false. Ferrario and her team had anticipated this, since models don’t have extremely high precision.  

From there, researchers measured cue count rates—the probability of song emission per hour—and calculated an average cue rate of 0.358 songs per group per hour. After accounting for cue rate, detection probability, and false positives, PAM produced an estimated density of 5.55 indri groups per square kilometer.

The known density of habituated groups in the same area was 6.17 groups per square kilometer.

This data revealed that the estimate provided by the PAM monitoring resulted in numbers close to reality, suggesting it could be a viable and low-cost technology for conservation efforts. “Every song we detect is a data point that could guide where conservation efforts need to go next,” says Ferrario. 

This successful PAM monitoring could have wider implications. Connor Wood, a conservation biologist at the Cornell Lab of Ornithology, who was not involved with the study, has seen PAM work for monitoring two species of owls, to almost a hundred species of birds, as well as wolves, coyotes, and amphibians. “The amount of species we can pull out of the same audio dataset really illustrates the power of bioacoustics for wildlife conservation,” he says. “That is a really powerful asset for conservation science.”

While the results and cost effectiveness of PAM monitoring are promising, it plays only one part in protecting the indri. 

The University of Turin and several partners have also supported Groupe d’Étude et de Recherche sur les Primates de Madagascar (GERP), a local association that manages the Maromizaha forest. The group helps locals grow native and endemic tree species, which are then planted in areas that have been affected by deforestation. The goal is to restore the forest and expanding suitable habitat for wildlife. Local guides and researchers take children out into the rainforest to observe animals like indris and teach them about ecology and conservation.

The hope is that one day, when the local children venture out into their home jungle, the air above them is even more musical.