How AI trained on birds is surfacing underwater mysteries

2026-02-09 · Source: The latest research from Google · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Bioacoustics · Depth: Intermediate, medium

Summary

Google DeepMind's Perch 2.0, a bioacoustics foundation model, demonstrates surprising efficacy in classifying underwater marine vocalizations, despite being trained primarily on bird and terrestrial animal sounds. Released in August 2025, Perch 2.0 was evaluated using a few-shot linear probe on marine tasks, including distinguishing baleen whale species and killer whale subpopulations, against models like Perch 1.0, SurfPerch, and a multispecies whale model. Utilizing datasets such as NOAA PIPAN, ReefSet, and DCLDE, Perch 2.0 consistently performed as the top or second-best model across various sample sizes, outperforming AVES-bird and AVES-bio on most tasks. This transfer learning capability is attributed to the model's large size, extensive training data, and its ability to learn detailed acoustic features from complex bird calls, which generalize well to other bioacoustics challenges.

Key takeaway

For bioacoustics researchers and marine conservationists aiming to classify new or elusive underwater sounds, Perch 2.0 offers a highly efficient transfer learning solution. You can leverage its pre-trained embeddings, even without underwater training data, to rapidly develop custom classifiers with minimal computational resources. Explore the provided Colab demo to implement this agile modeling workflow for your specific marine species monitoring needs, potentially accelerating discovery and conservation efforts.

Key insights

Terrestrial bioacoustics models can effectively transfer learning to underwater marine sound classification tasks.

Principles

• Larger models generalize better across diverse soundscapes.
• Detailed feature learning from complex sounds aids transferability.

Method

Apply a pre-trained bioacoustics model to generate embeddings from audio data, then train a simple logistic regression classifier on these embeddings for custom classification.

In practice

• Use Perch 2.0 for whale vocalization classification.
• Create custom classifiers with few labeled examples.
• Access NOAA data via Google Cloud for marine acoustics.

Topics

• Bioacoustics Foundation Models
• Transfer Learning
• Marine Bioacoustics
• Whale Vocalization
• Perch 2.0

Code references

• google-research/perch
• google-research/perch-hoplite
• earthspecies/aves

Best for: AI Researcher, Data Scientist, Machine Learning Engineer

Related on AIssential

• 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…
• How We Used Google's Perch v2 to Build a Bird-Based H5N1 Early Warning System in Patagonia — Bioacoustics foundation models can enable large-scale environmental monitoring for critical conservation challenges.
• 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.
• 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.
• Phonetic Error Analysis of Raw Waveform Acoustic Models — Raw waveform models with BLSTMs and transfer learning improve phone recognition, with errors reflecting inherent phonetic similarities.
• MOSS-Audio Technical Report — MOSS-Audio unifies audio understanding across speech, sound, and music using novel architectural and data strategies.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The latest research from Google.