How AI is letting scientists listen in on animal chatter
The fictional character of Dr. Dolittle has captured the imagination of millions of children with his ability to talk to animals – and now the idea of using technology to listen to and better understand animals is capturing the imagination of AI experts around the world.
For example, AI language-analysis technology is being used to decode the sounds of bottlenose dolphins and compile a dictionary of dolphin language.
This work is taking place on a global scale, across a vast variety of species. Researchers are using technology to gather data that could address some of the biggest environmental challenges of our time, including some with grants from Microsoft’s AI for Earth program helping achieve their goals.
Here is a snapshot of some of the projects underway – and what they hope to achieve.
[Subscribe to Microsoft on the Issues for more on the topics that matter most.]
Listening to elephants to aid conservation efforts
Acoustic monitoring methods are being used to identify and track the African forest elephant – a species whose numbers are in steep decline in the rainforests of Central Africa. The Elephant Listening Project uses 50 sensors to monitor almost 1,250km2 (485 miles2) of tropical forest and check in on the elephants as they roam around.
Conservation Metrics – which has recently been awarded a grant from Microsoft’s AI for Earth – is helping these listening projects scale their services through cutting-edge remote-sensing technology, machine-learning techniques and cloud-based analysis of data, making their conservation efforts even more effective.
The insect sounds revealing the health of the wider ecosystem
The mating call of the katydid (a relative of crickets and grasshoppers) may not seem very important in itself, but to scientists studying the health of Central America’s rainforest ecosystem, it could be key to their research. These insects occupy a central position in tropical food chains, and how they interact with both plants and animals can reveal valuable information about a wide variety of issues.
A team that includes researchers from Cornell’s Bioacoustics Research Program and Dartmouth University – with the help of an AI for Earth grant – is currently using a combination of acoustic sensors and AI to monitor a hundred varieties of katydids, and hopes to expand to other species, such as monkeys, in due course.
“Acoustics is a great tool to study many of the environmental issues we are facing today, but the analytics part has always been holding us back. These new AI techniques open up so much research we can do. It is a very exciting time for us,” explains project researcher Holger Klinck.
Using birdsong to predict biodiversity loss
Birdsong is both beautiful and highly complex. While seasoned birdwatchers can distinguish between different species and even different dialects of birdsong, most of us cannot. Enter the power of the computer.
Dr. Justin Kitzes, from the University of Pittsburgh, with help from an AI for Earth grant, is using machine-learning models to identify different types of recorded birdsong. This allows scientists to identify species that are in decline or are under threat and thus predict biodiversity loss more accurately.
“The data collection enabled by our work will ultimately help to form the foundation for effective conservation of birds and other vocal species on a changing planet,” says Dr. Kitzes.
Beyond Microsoft: Analyzing chicken squawks to improve animal welfare
There are efforts beyond the AI for Earth grantees in the listening space. Could you tell the difference between a stressed chicken and a content one? Scientists from the University of Georgia and Georgia Institute of Technology have been collecting the noises made by chickens in different circumstances – including the proximity of predators, air temperature or even the amount of mucus in a bird’s throat, which can indicate a viral infection.
The audio is then analyzed by a machine-learning program trained to recognize the different sounds coming from the birds. This data can then alert farmers to any issues the birds may have, enabling them to take immediate action if required.