As it turns out, the best way to track people who are cutting down trees is sound.
As it turns out, the best way to track people who are cutting down trees is sound.

What AI Hears in the Rainforest

As it turns out, the best way to track people who are cutting down trees is sound.

Topher White founded the nonprofit Rainforest Connection with the intent of creating a low-cost monitor that could help remote communities in their efforts to halt illegal logging, which is an enormous threat to tropical habitats. As it turns out, the best way to track people who are cutting down trees is sound. Using old cell phones linked to an artificial-intelligence platform in the cloud, White developed a system that can detect chainsaws in real time and send automated alerts to authorities. Today, Rainforest Connection is recording audio continuously from over a 1,000-square-miles of forest across 12 countries. That scale, along with rapid improvements in machine learning, have opened up tantalizing possibilities for understanding what the sounds of nature really mean.

Podcast Transcript

Editor’s Note: Transcriptions of episodes of the Outside Podcast are created with a mix of speech recognition software and human transcribers, and may contain some grammatical errors or slight deviations from the audio.




Outside Podcast Theme: From Outside Magazine and PRX, this is the Outside Podcast.

Michael Roberts (Host): There’s that age-old philosophical question: If a tree falls in the forest and no one is around to hear it, does it make a sound? The answer ultimately comes down to how you define sound. Is it the noise that something makes, or is it the act of hearing it? Now, here’s a very similar question that I can answer definitively. If an earthquake happens in the forest, and nobody is around to hear it, does it make a sound?

Yes it does. And it sounds like this.

[Audio clip of shaking leaves]

Roberts: Those are shaking branches, in a forest in Costa Rica. And as the tremors get bigger, they shake more. And get louder.

[Audio clip of shaking leaves continues]

Roberts: And then, as the shaking slows, something more interesting happens. A group of nearby spider monkeys reacts to the experience. And—despite being monkeys—they kind of go ape shit.

[Audio clip of monkeys hollering while leaves shake]

Roberts: This recording was captured by a group called the Rainforest Connection. They used a simple device, built from an old, repurposed cell phone.

Topher White: My name is Topher White, founder, CEO of Rainforest Connection, a nonprofit tech startup based here in San Francisco.

Roberts: White launched Rainforest Connection in 2013 with a mission to aid remote communities in their efforts to halt illegal logging, which is an enormous threat to tropical forests. The United Nations has estimated that organized crime accounts for 50 to 90 percent of the logging in equatorial regions. The loss of rainforest is also a major contributor to climate change.

One of the great challenges to stopping illegal logging is catching people in the act. We know that loggers tend to operate near roads. Satellite images and aerial photos can show where the forest is disappearing—but only after the fact.

It turns out that the best way to track people who are cutting down trees—is sound. Over the last seven years, White has engineered a simple, low-cost system that can detect illegal logging in forests across the globe in real-time—and then send automated alerts to local authorities. Along the way, he’s developed an artificial intelligence platform that offers tantalizing new possibilities for understanding what the sounds of nature really mean.

[Audio clip containing sounds of nature]


Roberts: If you’re going to get excited about forest sounds, there’s probably no better gateway drug than gibbons, those highly active members of the ape family known for singing duets.

[Audio clip containing gibbons singing duets]

Roberts: These are gibbons on the Indosesian island of Sumatra. Their song was captured by another Rainforest Connect device. But for White, it was the Siamang gibbons at the San Francisco Zoo that captivated him when he was a kid.

White: Uh, and you can hear the, uh, the Siamang from, like two parking lots over outside the zoo.

[Audio clip containing gibbons singing duets continues]

White: And when you get in there, not only are they making these amazingly fantastic noises, but they, you know, they swing around in ways that even monkeys can't do. They're just the coolest animals ever.

Roberts: As White grew up, he never stopped thinking about gibbons, even while he studied physics and computer science in college. And in 2011, while working at an alternative-energy organization that was focusing on nuclear fusion, he took a vacation to Indonesia, to volunteer at a gibbon reserve.

One day, he went for a walk in the forest with the head of the reserve and some rangers. Almost right away, they stumbled on illegal loggers cutting down a tree.

White: That struck me as kinda nuts that you could be a five minute walk from the ranger station with these sincere guards, uh, who weren't doing anything wrong, but the forest is noisy, and it's hard to really monitor this stuff on foot. Uh, and that struck me, especially with like a physics and science background, it's just not, not a problem that you can kind of let sit, you know?

Roberts: As an undergraduate, White had learned how computers can identify specific sounds within a larger soundscape. He’d also helped update his school’s old radio station, adding streaming capabilities as he learned to code.

Those experiences led him to believe that the solution to monitoring logging in the gibbon reserve was to listen for trouble. What he needed was a remote system to capture the noise of chainsaws and deliver it to the rangers.

Topher: And there was no, like road, there were no roads out there. There was no running water, there was no electricity, but there was cell phone service. Originally I thought we would build, I’d build some sort of specialized hardware, but the moment I started trying to do that, I was like, I can get old Android phones for five bucks on eBay.

Roberts: After White went home, he became increasingly obsessed with the concept. Eventually, he decided to quit his job and set up shop in his parents’ garage in San Francisco,  where, in 2012, it wasn’t hard to find talented tech people to lend a hand.

White: The cool thing about this type of project is that it captured people's imaginations and, and we got a lot of help. It wasn't for free, but, uh, I'd say it was discounted heavily.

Roberts: Still, designing a remote audio monitor that could be left out in the rainforest for extended periods was far more complicated than White expected. He knew he’d need to place the phones inside weather-proof boxes and outfit them with microphones. Beyond that, though, he was just going to have to figure it out.

White: But it turns out though, I took a lot of things for granted, like solar power. Like how to keep something reliable. Like how to get cell phone service?

Uh, I remember always thinking like, oh, we're going to be in the forest, we're going to have this thing powered by solar panels. We're going to, you know, it's going to be great. And, and then somebody having told me like, oh, you know, solar panels only work when there's like no shadow on them.

And I was like, oh, okay. Well, like maybe we're not using that much power, and walking through golden gate park in San Francisco into this like Redwood Grove and standing there and realizing that it was, like, dark under these trees and there was just no way that I was going to be able to get any solar power on the forest floor, uh, let alone potentially even at the top.

And it was just this moment of horror, you know, after having imagined how this whole system was gonna work, uh, and realizing we had to kind of start over from scratch on the solar side.

Roberts: Many of the tech people White consulted told him he was building his system all wrong. The obvious way to go about it, they said, was to configure the phones so they recognize the wine of a chainsaw—and nothing else. Instead, White had gone in the opposite direction, with the phones uploading every bit of the audio they captured to the cloud, where it would be analyzed by software. Sending all that data meant the monitors would need an order of magnitude more power.

He reached out to an innovative California solar company, and after a lot of trial and error, they worked out the power issue.

Then he turned to the biggest hurdle of all—the software. Here he had the luck of building the system just as artificial intelligence was becoming more sophisticated and more available. White pulled a number of data scientists into the project, and they adopted machine-learning tools that could sift through enormous amounts of audio to pick out a chainsaw.

Roughly two years after White first visited the gibbon reserve, he went back to Indonesia with a prototype that was… almost ready for action.

White: You know, I remember myself staying up all night with, uh, with like a net over me and like, you know, the biggest, uh, wasp possible try to get in, uh, just desperately trying to make sure that the software would work for the deployments we were supposed to do in the morning. It's really a miracle that it worked, um, at all.

Roberts: But it did work.

[Audio clip containing chainsaw noises]

Roberts: Shortly after the monitors were installed, White got an email alert. One of them had detected a chainsaw. The reserve rangers, with White nervously following, went into the forest and confronted the loggers, who put their saws down.

[Audio clip containing chainsaw noises abruptly stops]

Roberts: He had his proof of concept—though there was still a long way to go before he had a system that could make a lasting impact. The first units they tested in Indonesia? They barely lasted a month. And then they tried deploying more, in other parts of the world.

White: It's amazing how you can build something in Silicon Valley, you can take it to Indonesia, uh, to a rainforest and like miraculously it works, right? And you're like, wow, this is going to be great. And then you take it to Cameroon and you put it on a tree, it doesn't work at all. And you spend three, four months beating your head against the tree, trying to figure out how to make a technology work there that, you know? You used to work in, in Indonesia and Silicon Valley. And so every new place you've gone to, uh, that sort of thing would happen for a while until a little by little, we addressed the issues that were out there.

Roberts: In addition to all the technical difficulties, there were those unpredictable natural hazards.

White: Like in Peru, we put it up and within two days these termites came and deconstructed every piece of rubber, uh, and like non-rigid plastic on this, took it all back. You know, there must've been a million termites that came and took it all and why, who could say? Why are they attracted to one thing or the other? But they decided they wanted it, and so they took it. Uh, and you know, suddenly it's a new problem in a new forest that you have to address. And so that's this kind of like arms race that we're involved in all over the world.

Roberts: A critical piece of the system that White and the Rainforest Connection created is the Ranger App. When the artificial intelligence detects a threat in an area, it sends a push notification to local authorities and community members who have the app, telling them where the threat was detected, and letting them playback the sound so they can confirm for themselves, ‘Yeah, that’s a chainsaw.’

There’s just one problem with all this.

White: But it turns out that if you suggest to somebody that there's a chainsaw or a, or a vehicle in the background noise, they're going to hear it no matter what, even if it's there or not.

Roberts: For this reason, the app also delivers a spectrogram, which is a visual representation of sound. Users can zoom in on the image and analyze where the chainsaw stands out in the range of frequencies.

White: And so we can almost re-package the forest, uh, the sounds of the forest and deliver it to them in a way that they're able to verify and come to understand it so much better.

But on top of that, it's also just part of this, this, uh, camaraderie—part of this collaboration with them. You don't want, you don't want to be telling people what to do. You want to be saying, ‘look what we think is true. Uh, if you think it's true, you get out there.’ It really, uh, underscores this idea that we work for them.

Roberts: Still, it’s one thing to effectively detect that illegal logging is taking place. Stopping it? There’s no app for that.  

White: There's this assumption in the work that we do, and some of this came from the first time we tested it, that people who say that they want to get real time alerts, that they'll show up and actually stop the loggers. But that’s a huge open question. You show up, and you stop a logger with a chainsaw, and potentially with weapons and a truck and all the rest. What are you supposed to do then?

Roberts: The answers vary by location, and none are perfect. In Peru, Rainforest Connection partnered with a group of lawyers, giving them access to the Ranger App so they got the alerts at the same time as local tribes protecting the forest. The tribes would apprehend the loggers and bring them to the police, and the lawyers would file a case at the same time. It’s made a big difference.

New features to the Ranger App have also helped. Location awareness allows rangers in an area to collaborate better. And they can upload geo-tagged photos that can be used as verified evidence. Of course, none of this will solve the issue that, in many places, law enforcement is unable or unwilling to take on loggers.

White: It's still a challenge for us to figure out like, what's the next step? What's the next step? And you know, now we think that we're close to closing that loop. It might be that we're only 25% of the way towards, uh, closing that loop. Maybe it becomes a more and more complicated process as we go, but that would be in line with the way conservation often is.

Roberts: Today, Rainforest Connection has monitors at more than 20 sites in 12 countries. Each unit can capture sounds up to roughly a mile away. Add it all up, and they are recording audio continuously from over 1,000 square miles of forest.

That scale gave White a new idea. What, he wondered, if we started using artificial intelligence to listen for more than just chainsaws?

[Audio of rainforest noises]

Roberts: We’ll be right back.




Roberts: White’s choice to have the Rainforest Connection monitors upload all the sound they were capturing may have been the “wrong” approach according to a lot of tech engineers, but it has ended up putting him in a position to have a much broader impact than he ever imagined.

White: We built a system that was essentially streaming live audio from the forest, all these really remote places that nobody could go to to our server. And yes, we were looking for threats, but we could also turn that audio around and, you know, make it available to people.

Roberts: Early on, White had launched a Kickstarter campaign to fund his project. He figured that one of the best ways to get people interested in the idea was to offer them the chance to listen in on the places where Rainforest Connection was working. So as part of the campaign, he proposed a mobile app that would do exactly that.

White: So, when we sort of launched this Kickstarter campaign, it was like, ‘hey, do you want to hear what’s happening in Sumatra? In Africa? In South America? Do you want to hear that live?

Do you want to have this connection to the forest?’ And people said, yes, that sounds cool

Roberts: It is pretty cool. Open the Rainforest Connection app at the right moment, and you might catch a troop of howler monkeys.

[Audio containing howler monkeys]

Roberts: Or you might even hear a chainsaw. It happens.

White: And even though we build this for people to be able to connect to the joy of the forest, there have been well over a dozen times where I will get a text message or an email from a person I don't know, who says, ‘I'm listening to this stream. There's a chainsaw there.’ And I'll tune in.

[Audio containing chainsaw noises]

White: And I'll, and I'll say, yep, there's a chainsaw there. The alert, the rangers got an alert, but they were doing something else. And so then, suddenly we can turn our attention to the rangers and be like, ‘what’s up?’, and they’re like, ‘oh wow, sorry, we were doing something else right now.’ So, it’s this, suddenly you’ve turned these, uh, these precious places into, not just the, um, the care and responsibility of the people on the ground, but the care and responsibility of the whole world.

Roberts: Listening to the cacophony of life can make people feel responsible for it, and to want to take action to protect it.

But the potential value of the system that he’s created goes much further. For scientists in the relatively young field of bioacoustics, a network of always-on remote monitors that are feeding audio into an A.I. platform—that’s a huge leap forward.

Until relatively recently, studying an ecosystem through sound meant going into the field to record, and then the daunting task of listening to days or weeks worth of tape to pick out the species you’re trying to understand. This is the kind of big data task that machines can do much faster—and better than we can.

White: The way that nature interacts with itself is, uh, is often through sound. And that's something that we're not very good at picking out, especially because it’s temporal. You can’t really absorb a week’s worth of sound in the same way that you can absorb a week’s worth of video, or potentially even, um, images.

[Audio containing various rainforest noises]

White: Maybe you hear a car drive by, maybe you hear gibbons, maybe you hear monkeys, maybe you hear a half a dozen different birds. You can listen to the audio and potentially pick out one of those things or two of those things. But a computer could pick out all of those things continuously over time. And then look for correlations between what's happening in nature any given time. Even the rare, rare things.

Roberts: Not surprisingly, big tech has taken an interest in White’s work. Over the last couple years, he’s received funding from Google while using their open-source A.I. platform, TensorFlow. Wah-Weh, the Chinese technology giant that’s under fire from the Trump Administration approached White with a particularly ambitious proposition after learning about a project that was using Rainforest Connection monitors to track spider monkeys in Costa Rica.

White: Detecting where spider monkeys are, was not a, um, an intense enough problem for them. So they said, you know what, let's understand SpiderMonkey language. Um, so, so they set out to do that. Uh, the early version of this turned out to be a sentiment, uh, analysis, meaning like, are spider monkeys upset? Are they happy? Are they calm?

Roberts: Artificial Intelligence isn’t able to understand monkey language yet. Maybe it never will. But the way White and his team designed their system, it should continually get better at interpreting the sounds of the forest.

The platform is modular, meaning scientists can add new algorithms that listen for very specific vocalizations of select creatures. Imagine hundreds of algorithms analyzing the same piece of audio to reveal an increasingly intimate portrait of an ecosystem.

White believes it’s even possible for A.I. to hear things that don’t make any sound at all. Like, say, a big cat prowling the forest.

[Continued audio containing rainforest noises]

White: It's not going to always make a noise. It's not going to be roaring. Uh, in fact, in many ways, uh, we wouldn't hear it at all. But you can be sure the birds are talking about it. Uh, and that's even true for a tiger or a jaguar or a person with a gun. We're not going to be able to detect the rustling of leaves when a person steps on them. But you can definitely detect when a bird is putting out a warning call.

[Audio containing bird noises]

Roberts: If you really want to know what all this means for our understanding of the chorus of nature, you need to talk to Bernie Krause. An early pioneer in the field of soundscape ecology, he has been recording in various habitats around the world since the late-1960s.

Among his influential contributions is the niche hypothesis, which holds that each organism in a wild soundscape has evolved to make vocalizations within a specific frequency or a specific time of day, or both. Put another way, if a creature wants to be heard, it needs its own sound, or its own time to make sound.

[Audio of Krause’s Kenya recording containing various animal noises]

Roberts: Krause captured this recording in Kenya, in 1983, at a place called Governor’s Camp. Listen closely, and you can pick out insects—frogs, elephants eating something, and the eerie wailing of a hyena.

[Audio of Krause’s Kenya recording continues]

Roberts: The audio quality here is incredible. This is partly because Krause had been a professional musician before turning to science. He and his late music partner, Paul Beaver, introduced the synthesizer to popular music and film. But Krause also believes that if you want to tell the story of a place through sound, you have to listen very carefully, with meticulously calibrated equipment.

Bernie Krause: To me these soundscapes have real meaning. They’re narratives of place and if you get the right kinds of carefully calibrated field data, recorded with dedicated protocols, uh, you can actually both hear and see the effect on wildlife that result from the machinery that cuts down forests or ravages the earth with drilling or open pit mining. You can actually hear the differences between a healthy habitat and one under stress.

Roberts: Krause made headlines in 2015 when he released recordings and a spectrogram that presented the dramatic change in the Northern California soundscape during the recent multi-year drought in the West. All of them were captured in the spring, in the exact same location, at a site in Sugarloaf Ridge State Park.

[2004 audio of Sugarloaf Ridge State Park containing sounds from a vibrant community of birds]

Roberts: In 2004, you can hear what sounds like a rich and vibrant community of birds. 

[2009 audio of Sugarloaf Ridge State Park, containing some bird noises.]

Roberts: In 2009, there’s slightly less activity.

[2015 audio of Sugarloaf Ridge State Park, containing barely any bird noises.]

Krause: And at one point in 2015, it became a silent spring. But it was remarkable how haunting the soundscape was at that particular moment

Roberts: Krause calculated that the totality of sounds produced by living organisms at Sugarloaf dropped by a factor of five between 2004 and 2015.  He told me that he expects that this kind of precise study of an environment over time is close to impossible if you’re using recycled cell phones, because they can’t be calibrated to give you apples to apples comparisons.

I asked Topher White about this and he said that’s absolutely true—which is why he’s working on new monitors that will have superior audio recorders and just use the old phones to transfer the data.

Anyway, none of this is to say that Krause is a skeptic of Rainforest Connection. He is a huge fan of the monitoring project, and he sees enormous potential in artificial intelligence that could offer new insights to his archive of some 5,000 hours of habitat recordings, which he calls the Wild Sanctuary.

On a very personal level, though, Krause just isn’t so into the idea of letting technology do all the listening.

Krause: What's the point of setting up all of this, um, remote monitoring equipment, if you're not outside? Because you know, when we're recording, I don't care what kind of technology we use, uh, to capture the sound. The result is an abstraction of what's out there. It's a very small part of the whole habitat, no matter what mikes we use, no matter what equipment we use. I don't have the same relationship to a piece of a, to a recording that I've made when I set a recorder up and walk away for a whole night and go to sleep in my tent and wake up the next morning and collect the data. That whole period of time that occurred that night is lost for me. Um, most of the experience is lost for me. But that's just me personally. I happen to like being there when things occur.

Roberts: Where Krause and White seem to align most closely is in their belief that the sounds of nature must be shared with as many people as possible. This, they agree, is the only way to spur the kind of mass response needed to head off the crisis the planet is facing.

White achieves this through live streaming of the forest through the Rainforest Connection app. Krause, who is now in his early 80s, has turned back to where he started his audio career—the arts.

Krause: The only way to get to large groups of people is through the arts. What I've done is, uh, I've taken my work in my archive, and I'm beginning to transform it into works of art.

[Audio containing a bird cooing]

Roberts: The most powerful result of this so far has been the “The Great Animal Orchestra” an immersive museum exhibition.

[Audio containing wolves howling plays beneath narration]

Roberts: The centerpiece, created in collaboration with United Visual Artists, a London-based multi-media collective, is a pitch black room, where interwoven spectrograms light up the walls, transforming a selection of Krause’s recordings—including these wolves in Ontario’s Algonquin Park— into an an almost hypnotic sensual experience.

[Audio containing wolves howling continues. Birds chime in as well.]

Roberts: The installation has gone up in Paris, Shanghai, Milan, Seoul, and London. It has been experienced by more than a million people.

[Audio containing wolves howling slowly fades.]

Roberts: Bernie Krause’s website is He shared all the incredible habitat recordings in the second half of this episode.

Topher White is a 2019 recipient of a Rolex Awards for Enterprise, which recognizes trailblazing scientists and innovators from around the world. You can learn more about his work at The Rainforest Connection app is easily found in the Apple App Store and on Google Play.

This episode was produced by me, Michael Roberts, with music by Robbie Carver.

This episode is brought to you by Visit Arizona, home of world-renowned artist Chip Thomas. To learn more about his Painted Desert project and what makes Arizona so alluring to artists, go to

We’ll be back next week.

Follow the Outside Podcast

Outside’s longstanding literary storytelling tradition comes to life in audio with features that will both entertain and inform listeners. We launched in March 2016 with our first series, Science of Survival, and have since expanded our show to offer a range of story formats, including reports from our correspondents in the field and interviews with the biggest figures in sports, adventure, and the outdoors.