In Southeast Asia's tropical forests, a chorus of calls rings through the canopy long before any human observer arrives—and Dr. Sean Yap is listening to every note. As a Research Fellow at the National University of Singapore's Center for Nature-based Climate Solutions, Yap has discovered that the soundscapes of tropical forests hold secrets to biodiversity that traditional field surveys can never fully capture. His team has combined bioacoustics—the study of animal sounds—with artificial intelligence to transform continuous audio recordings into tools for ecological understanding, monitoring everything from the impact of traffic noise on wildlife to the promise of tiny restored forest patches in cities.
The challenge that makes this work urgent is one of scale and consistency. Tropical forests in Southeast Asia are exceptionally dense and diverse, making it nearly impossible for researchers to spend sustained time surveying species by conventional means. A human observer, no matter how skilled, will miss creatures that avoid their presence and can only document what occurs during limited fieldwork hours. The recorders Yap's team deploys are different: small autonomous devices with microphones tuned to specific frequencies that can collect data continuously or at programmed intervals, capturing standardized datasets night and day without the limitations of human fatigue or observer bias.
Once collected, the audio undergoes transformation. The recordings are converted into spectrograms—visual representations of sound—which AI systems then analyze to identify patterns. Some models distinguish broad categories like traffic noise versus animal calls; others classify bird vocalizations to species level or differentiate between anthropogenic sounds and wildlife. This capability allows researchers to assess how human disturbance affects animal activity and to measure the presence and behavior of different species across study areas with far greater precision than traditional surveys could achieve.
Yet the technology is not without constraints. Yap's research has revealed that AI models perform well for species with distinctive vocalizations—songbirds, for instance—but struggle with lower-frequency calls from pigeons, doves, and owls. Early results showed recordings flagged as owl calls that turned out to be traffic noise. The deeper problem is one of geographic bias: many existing sound-recognition models are trained primarily on North American and European species, making them less reliable for Southeast Asian wildlife. To address this, NUS researchers are refining models using locally collected data, recognizing that as regional biodiversity datasets expand, the AI systems will grow more accurate.
What makes this work compelling is not the technology itself but what it enables. Bioacoustics transforms how scientists gather information about dense, biodiverse ecosystems where traditional methods fall short. Yap is currently exploring two pressing questions: how human-generated noise affects animal activity in forests, and whether microforests—small restored patches in urban areas—can improve ecological connectivity. These are not abstract concerns. They speak directly to how cities coexist with nature and how fragmented landscapes might be mended.
Yap is careful to emphasize that AI complements rather than replaces field expertise. "AI tools depend on good training data and ecological expertise," he notes. The future of conservation in biodiversity-rich regions like Southeast Asia may well depend on this partnership: machines listening where humans cannot linger, but scientists providing the knowledge to interpret what they hear.