How We Used Google's Perch v2 to Build a Bird-Based H5N1 Early Warning System in Patagonia

2026-04-29 · Source: HackerNoon · Field: Science & Research — Environmental Science & Earth Systems, Life Sciences & Biology, Artificial Intelligence & Machine Learning · Depth: Intermediate, medium

Summary

Fundación Kreen, a Chilean Patagonian conservation NGO, collaborated with Google Research to conduct the first complete acoustic inventory of the Meullín-Puye Nature Sanctuary. Using Google DeepMind's Perch v2 bioacoustics foundation model, which covers 14,795 species, they analyzed 4,998 FLAC recordings from three Bird Weather PUC acoustic sensors. This effort, dubbed "Silent Vectors," aims to monitor migratory seabirds as potential vectors for antimicrobial resistant bacteria and H5N1 avian influenza, which is a growing concern in Chile's salmon farming regions. The project yielded 49,383 acoustic detections across 35 species, including 1,343 detections of the Vulnerable Flightless Steamer Duck, a critical sentinel species previously absent from Perch 1's training data. The team also received direct technical support from a Google Research scientist, which helped resolve critical implementation issues.

Key takeaway

For AI Scientists or Conservation Directors developing biosurveillance systems, this project demonstrates how foundation models like Perch v2, combined with strategic geofencing and expert support, can rapidly establish critical baselines for environmental health. You should consider integrating open-source bioacoustics models into your monitoring protocols to scale data collection and address urgent ecological threats like H5N1 and antimicrobial resistance, especially in remote or rapidly changing environments.

Key insights

Bioacoustics foundation models can enable large-scale environmental monitoring for critical conservation challenges.

Principles

• Geofencing improves acoustic model signal quality.
• Open-source models enable scientific reproducibility.

Method

Acoustic sensors record ambient audio, which is then processed by a bioacoustics foundation model with geofencing and calibrated thresholds to detect and identify species for environmental monitoring.

In practice

• Use `perch-hoplite` to load Perch v2.
• Update TensorFlow to 2.20+ for Perch v2.
• Download Perch v2 labels separately.

Topics

• Google Perch v2
• H5N1 Avian Influenza
• Bioacoustics Monitoring
• Chilean Patagonia
• Salmon Aquaculture

Best for: Research Scientist, AI Scientist, Director of AI/ML

Related on AIssential

• How AI trained on birds is surfacing underwater mysteries — Terrestrial bioacoustics models can effectively transfer learning to underwater marine sound classification tasks.
• A fully automated framework for acoustic identification and localization of terrestrial wildlife at scale — Automated acoustic localization using low-cost hardware and open-source software enables large-scale wildlife monitoring.
• GetNetUPAM: Ecologically Informed Nested Cross-Validation and Noise-Robust Attention for Marine Bioacoustic Monitoring — Ecologically informed nested cross-validation and adaptive attention improve marine bioacoustic model stability and generalization.
• AI Trained on Birdsong Can Recognize Whale Calls — Google DeepMind's Perch 2.0, an AI audio model primarily trained on millions of land-based animal recordings including birds, amphibians, insects, and mammals, has shown unexpected strong performance…
• A finely annotated dataset for the automated acoustic identification of European Orthoptera and Cicadidae — A new dataset combines weak and strong acoustic labels for automated European insect identification.
• Dolph2Vec: Self-Supervised Representations of Dolphin Vocalizations — Dolph2Vec, a species-specific SSL model, effectively uncovers fine-grained dolphin communication patterns using a novel, longitudinal dataset.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by HackerNoon.