It was amid the stillness of the Covid-19 lockdowns that Conrad Young, founder and CEO of Chirrup.ai, realized that birdsong would make an excellent proxy for biodiversity. Birds have always acted as sentinels for changes in the environment, and as he told EE Times, “Birdsong is a pretty powerful and visceral way of understanding what’s going on in the environment but is also scientifically valid and benchmarkable.”
In March 2022, as the pandemic ended, Young launched Chirrup.ai to monitor birdsong and track farmland diversity. The first task was to deploy acoustic sensors and establish a baseline of bird activity that could provide a reference point against future recordings and reveal whether biodiversity was improving, declining, or remaining stable over time. Earlier trials used audiomoths—low-cost, acoustic loggers that record birdsong—but Chirrup.ai has now developed smart acoustic sensors.
The sensor pairs a MEMS analog microphone with a 24-bit digital audio codec, a 240-MHz processor with Bluetooth, and a high-speed 4G LTE modem. A multi-constellation GNSS receiver supporting GPS, GLONASS, and Galileo provides geolocation and timestamping for the acoustic data. The device is powered by a rechargeable lithium-polymer battery.
Chirrup.ai’s technology has already measured biodiversity on more than 300 farms in eight European countries. In the U.K., these include sites on the Greenwich Peninsula in London, broiler farms in Lincolnshire, and pig farms in Norfolk. To record birdsong in the field, multiple devices are placed at least 200 meters apart at the edges of woodland and open land populated by birds but away from roosting sites, footpaths, and roads. Birdsong is recorded for two to three weeks with devices then sent back to Chirrup.ai, where the recordings are converted to spectrograms and analyzed to identify the species and assess biodiversity.
View All As Young pointed out, farmers do not require instant answers when assessing biodiversity but instead want to understand how nature is changing year on year. “We need a large enough period of sampling, so we’ve got a good representative picture of what’s there, and then we’ll provide an annual snapshot,” he said.
Chirrup.ai technology director Paul Nebel agreed and added that remote analysis of data is preferable to using edge AI right now. “With edge processing, we would need a bigger battery and probably a solar cell, which would make the device way more expensive,” he said.
Sensor aside, identifying bird species from recordings is critical, and Chirrup.ai has developed machine-learning (ML) algorithms to detect subtle nuances in recordings, such as one species mimicking another, and remove background noise. The firm also uses ML platform BirdNET, which is trained on species worldwide.
“Birds are very individual: Some use the same songs day in and day out, but others do not and are very complicated,” Nebel said. “Many have songs to raise an alarm, for waking up, for saying hi to their mates, and for courting. Some birds pretend to be other birds; others even pretend to be mobile phones. We also see birds that have different accents from one country to another, even from county to county.”
These remarkable complexities make human oversight critical, and Chirrup.ai has been working with ornithologists and ecologists who have reviewed thousands of hours of recordings, comparing their judgements against ML predictions, and setting appropriate confidence thresholds. “We need to keep checking that the machine learning doesn’t drift off and remains reflective of what humans hear,” Nebel said.
Chirrup.ai has also developed algorithms that generate a biodiversity index based on its bird species data as well as local environment conditions, such as hedgerow health, land management practices, satellite data on habitats, and species position in the food chain. Work is underway to include other bioindicators, such as weather data, to contextualize changes from season to season, and the firm is also considering extending bioacoustic sensing beyond birdsong. “There are also so many nascent datasets that are becoming available as well as birdsong—we could look at insects, mammals, even the livestock,” Young said.
One firm already working with insects and, very recently, livestock is AgriSound (U.K.). Founded in 2020 by CEO Casey Woodward, the government-funded startup set out to monitor pollinating insects to help growers improve crop pollination and provide insights into biodiversity for farming and land management. Woodward and colleagues developed a solar-powered acoustic sensor, Polly, that monitors honeybee and bumblebee activity in glasshouses, polytunnels, and open fields by detecting wingbeat frequencies.
The sensor combines a MEMS microphone and acoustic sensor array to capture the insect wingbeat sounds, alongside temperature, humidity, and ambient-light sensors to monitor environmental conditions.
An onboard processor runs ML algorithms to analyze the discrete Hertz range of the wingbeat frequency and distinguish bee activity from background noise. The resulting pollinator data packets are transmitted to the cloud via 4G cellular connectivity for visualization and interpretation, typically on an hourly basis, and a Bluetooth module enables firmware updates.
So while Chirrup.ai, with its emphasis on biodiversity insights, currently prioritizes battery life and cost by analyzing recordings remotely, AgriSound processes signals locally to deliver swift results on insect activity. “We are trying to help farmers who want results quickly,” Woodward said. “For example, the almond flowering season may only be 10 days, so growers will want to take recommended interventions pretty quickly.”
Like Chirrup.ai, AgriSound has spent some time developing and training ML algorithms to recognize, in this case, different pollinators. Woodward and colleagues have built an extensive library comprising thousands of hours of insect sounds, recorded under a controlled environment. Here, individual bees have been encouraged to fly toward a flower and collect pollen and nectar while their acoustic signatures are recorded. “Bees are very smart and difficult to incentivize to do what you want,” Woodward laughed.
The sensor has been deployed by U.K. retailer Marks&Spencer to provide insights into pollinator densities and can be found in strawberry, sugar beet, and other farms across the nation. Woodward highlighted how his company’s work in glasshouses has helped to reduce the amount of misshapen fruit. AgriSound’s tech has also reached almond orchards in Spain and Portugal, blueberry crops in the U.S., and avocado farms in South Africa—and is now being extended to livestock. In recent U.K. projects, Woodward and colleagues have worked with veterinary-led sustainable livestock production group Raft Solutions and egg producer Noble Foods to monitor the welfare of laying hens.
“With Raft, we’ve been listening for changes in the gut contractions [of cattle] and how these affect methane emissions from the herds,” Woodward said. “This can take time, as you need a veterinarian to collect the sounds with a stethoscope … but with the hens, we just put our non-invasive sensors into the sheds to record data.”
Pest control
Beyond birds, bees, and bovine, the next acoustic sensing frontier is, arguably, the smaller insect pests that infest crops and devastate yields—think beetles, termites, stem borers, and more. U.K.-based BugBug—launched by Cornell entomologist Emily Bick, director of Indonesian PT Gunung Madu sugarcane plantation Jadey Huray, and University of Wisconsin–Madison researcher and ML engineer Laurence Still—is tackling this problem head on.
Laurence Still, BugBug’s CTO, highlighted how a few years ago, Gunung Madu switched from burning fields before harvesting to the more sustainable method of green cane harvesting—that is, harvesting sugarcane without first burning the crop. “[The plantation] went from 5% crop loss to 50% crop loss due to insects over the course of a year,” Still said. “Burning resets the insect population back to zero, but without this, insect numbers were compounded and compounded.”
To tackle the problem, the colleagues decided to use piezoelectric contact microphones to detect insect vibration frequencies, connecting these to Raspberry Pi processors and clipping them onto plant stems to detect the vibrations caused by the chewing and tapping of feeding insects.
“We record the sound of the damage being done to the plant by the insect,” Still said. “A lot of biological acoustic sensing detects sounds transferred through air, but we’re listening to [vibrations] being transmitted through the plant. These travel very well through hard materials, so we can pick up very, very small signals.”
Trials on tobacco-plant-feeding hornworms, Colorado potato beetles, and European corn borers revealed distinct differences in the feeding patterns, frequency, and amplitude—and low background noise. Following initial success, the BugBug team, led by systems integration engineer Callum Hockey, has since redesigned its device, sticking with the piezo-sensors but switching the Raspberry Pi processor for a custom sound card and embedded electronics. This latest design decreases the form factor, cuts power consumption, and enables field operation with a smaller solar panel.
Recorded data is converted into spectrograms and analyzed using custom ML algorithms. Unlike birdsong and bees, the sounds of insects feeding internally cannot readily be identified and labeled, so BugBug’s ML approach relies heavily on unsupervised learning. Like their peers at Chirrup.ai and AgriSound, Still and colleagues have collected thousands of hours of recordings. These are combined with contextual information, such as insect location on the plant, and plant location, to label populations of sounds rather than individual events. By comparing these datasets, the ML algorithms learn to distinguish insect activity from background noise and identify patterns associated with particular pests. Some processing takes place locally while the Wi-Fi-enabled sensor also transmits data to the cloud for further analysis. However, BugBug is also exploring lightweight ML models that could run on the device in the future.
Still and colleagues have characterized the sounds of more than 25 insect species, amassing data on sugarcane, sweetcorn, oilseed rape, grapevine, coconuts, potatoes, and tomatoes. The sensors can also detect the distinctive feeding sounds of piercing-sucking insects, including aphids and scale insects—the scourge of sugarcane. It’s early days, but the company is investigating if acoustic monitoring can detect the first flight of these insects, just after they emerge from larvae, to give growers an early warning of pest activity.
“Our idea is to put the sensors in the field at the start of the season and build up a picture of insect activity over time,” Still said. “So you see no insects, no insects, no insects, and then you start to see your signal rising—and that’s the time to act. Identifying the optimal time to treat a crop is our goal—timing is absolutely critical.”
Practical proliferation
BugBug is undoubtedly breaking new acoustic sensing ground by using AI to uncover data that was previously inaccessible to humans. “We’re doing something [with AI] that wouldn’t have been possible five years ago, before some of the newer advances in unsupervised learning,” Still said. The company is currently working with leading agrochemical companies to solve practical problems. “You might have a company that has developed a biopesticide but needs to know if it is stopping insects feeding, killing them, and how quickly,” he said. “That’s hard if you can’t see the insect as it’s inside the plant … but with our sensors, you could monitor feeding activity before pesticide and post-pesticide.”
BugBug, AgriSound, and Chirrup.ai are not alone in their development of acoustic sensors and AI to eavesdrop on the environment. In Sri Lanka, Permia Sensing is combining satellite, drone, and bioacoustic data with AI to monitor red palm weevil activity in palm trees. Meanwhile, Biofonic (U.K.) uses below-earth acoustic sensors and ML to gather soil data for sustainable crop management.
Altogether, these companies point to a growing shift toward bioacoustic sensing and AI-driven analysis to check in on nature. Indeed, Young highlighted how the “amalgamation” and AI analysis of so much sound data with myriad databases, combined with years of work with growers, is truly helping agri-industry balance commercial production with ecology.
Woodward concurred. “The beauty of acoustic sensors is that they make the invisible visible,” he said. “To some extent, pest damage can be monitored by looking at bites on a leaf, but you can’t necessarily quantify the number of beneficial insects in a field or poor pollination. I see acoustic sensing as a statistically rich and powerful means of data collection, which in many cases is super-affordable.”
Read also:
[EE Times Magazine – November 2026](https://www.eetimes.com/ee-times-magazine-november-2026/)
[What It Takes to Let a Farm Robot Work Alone](https://www.eetimes.com/what-it-takes-to-let-a-farm-robot-work-alone/)
[Can Machines See a Sick Plant Before a Farmer Can?](https://www.eetimes.com/can-machines-see-a-sick-plant-before-a-farmer-can/)