AI Trained on Birdsong Can Recognize Whale Calls

2026-03-17 · Source: IEEE Spectrum · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Life Sciences & Biology · Depth: Advanced, quick

Summary

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 in classifying whale vocalizations. This success is attributed to transfer learning, enabling the model to apply knowledge gained from avian calls to cetacean sounds, thereby reducing computational time and experimentation effort. Researchers evaluated Perch 2.0 on marine audio datasets by converting sounds into spectrograms, generating embeddings, and training a logistic regression classifier, with results presented at a NeurIPS workshop demonstrating good performance even with limited data. The model's effectiveness is theorized to arise from evolutionary parallels in vocal production, the "laws of scale" for large foundation models, and its ability to recognize fine-grained acoustic characteristics across diverse soundscapes. This breakthrough offers a powerful tool for passive acoustic monitoring and aiding whale conservation efforts.

Key takeaway

Google DeepMind's Perch 2.0, an AI audio model trained on land animal bioacoustics, demonstrates strong transfer learning capabilities for classifying whale vocalizations. It achieves robust performance on marine datasets using a logistic regression classifier trained on as few as 4-32 embeddings, often outperforming or matching specialized models. This significantly reduces computation and experimentation effort, enabling scalable bioacoustic monitoring for marine conservation and discovery of new underwater sounds.

Topics

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

Best for: Research Scientist, AI Researcher, AI Scientist, Data Scientist

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.
• 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.
• 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.
• Speech intelligence for enterprises - Mistral AI — Mistral AI offers open, customizable speech intelligence models for enterprises, enabling human-like voice interactions and accurate transcription.
• 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 IEEE Spectrum.