Seema Lokhandwala captures elephant sounds at Kaziranga National Park. Credit: Vijay Bedi

Rumble, roar, trumpet, and chirp. Armed with artificial intelligence-powered algorithms, Seema Lokhandwala listens closely to Asian elephant sounds in the forests of Assam, a hotspot of human-elephant conflicts in northeast India.

She quickly interprets vocal sounds of captive and wild elephants to reduce such conflicts. "If we know which elephant is vocalizing, we can know if that animal is in conflict," says Lokhandwala.

In the Elephant Acoustics Project she is part of, AI and machine learning (ML) have been useful tools. The algorithms help her clarify overlapping sounds and decipher infrasonic ones inaudible to humans. They also separate trumpet calls based on whether the elephants are interacting with their mahouts or with other elephants.

Listening in to these calls, a researcher could take up to 30 minutes to decode the message. Lokhandwala says AI can figure out a hundred calls in five minutes.

Five years ago, ecologist V. V. Robin initiated a project to understand why birds found in some Western Ghats habitats did not appear in others. It took him two years to analyse avian sound recordings collected over a year. “With AI we would have completed the study in a year,” says Robin at the Indian Institute of Science Education and Research (IISER), Tirupati.

Lokhandwala, Robin and other researchers are putting to use the combined power of bioacoustics and AI to observe and conserve species and ecosystems. Alongside best use cases, they are making a note of the challenges with AI and building high-quality training datasets.

"We analyse thousands of samples using acoustics,” says entomologist Devi Shankar Suman at the Zoological Survey of India. Their recent study1 deciphered buzzing mosquitoes to detect species, sex, and feeding status.

Using RavenPro, a sound analysis software developed by Cornell University, the team homed in on 19 mosquito species, including those that cause dengue and malaria. Suman and colleagues will use this data to build an AI-integrated mobile application for surveillance of real-time vector-borne diseases.

At the 86-acre campus of Bengaluru's Srishti Manipal Institute of Art Design and Technology, an array of 20 recorders pick up sounds. Each recorder collects 144 minutes of data every day creating a reference dataset that can point to changes in the landscape and biodiversity. The AI models are trained on sounds of collective acoustic signatures of organisms (biophony), of the environment (geophony) and humans (anthrophony), says Gururaja K V, a batrachologist at the institute.

These AI-driven methods can provide continuous, real-time surveillance without the need for human intervention. "Real-time AI analysis could alert authorities to activities such as illegal tree cutting."

Vijay Ramesh records bird songs in the Anamalai hills. Credit: Priyanka Hari Haran

Extensive training datasets are needed to better monitor under explored taxonomic groups so that robust automated recognition algorithms can be produced. "Compared to birds, we have limited audio data on other taxa like amphibians and insects," says Vijay Ramesh, a postdoctoral research fellow at K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology, in the United States.

Burooj Ghani, a postdoctoral fellow in AI and biodiversity at the Naturalis Biodiversity Center in the Netherlands, has demonstrated that bird sound models can be repurposed as pre-trained models for other taxonomies with limited data.

Initiatives such as Soundscapes to Landscapes in Sonoma County, California, tap into machine learning-based bioacoustics monitoring of bird sounds across large areas. Canada's SWAG (Ships, Whales, Acoustics in Gitga'at Territory) project harnesses AI to identify marine species and understand the nature of the vocalization in Gitga'at Territory.

Robin says birds from India need to be better represented in AI-enabled platforms such as BirdNET, which identify birds through sounds. The algorithm represents most North American and European birds. "Many from India are not included," he says.