Perch 2.0: The Bittern Lesson for Bioacoustics

2025-08-06

AI Summary

To address insufficient cross-species (beyond birds) sound classification and transfer learning capabilities in bioacoustics, this paper introduces the first large-scale self-distillation pretraining framework targeting multiple taxonomic groups. Methodologically, it innovatively incorporates fine-grained species classification into the self-distillation pipeline, jointly optimizing a prototype-based classifier and a source-prediction objective to enhance supervised signal quality and learn domain-invariant representations. The proposed model achieves state-of-the-art performance on the BirdSet and BEANS benchmarks and significantly outperforms specialized models in zero-shot and few-shot marine bioacoustic transfer tasks—demonstrating strong generalization and cross-domain adaptability. Key contributions include: (i) establishing the first multi-species self-distillation paradigm; (ii) proposing a prototype-source joint learning mechanism; and (iii) enabling efficient bioacoustic representation transfer without requiring any target-domain data.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Perch is a performant pre-trained model for bioacoustics. It was trained in supervised fashion, providing both off-the-shelf classification scores for thousands of vocalizing species as well as strong embeddings for transfer learning. In this new release, Perch 2.0, we expand from training exclusively on avian species to a large multi-taxa dataset. The model is trained with self-distillation using a prototype-learning classifier as well as a new source-prediction training criterion. Perch 2.0 obtains state-of-the-art performance on the BirdSet and BEANS benchmarks. It also outperforms specialized marine models on marine transfer learning tasks, despite having almost no marine training data. We present hypotheses as to why fine-grained species classification is a particularly robust pre-training task for bioacoustics.

Problem

• Expands bioacoustic model to multi-taxa from avian-only training
• Improves species classification via self-distillation and prototype-learning
• Achieves state-of-the-art performance on BirdSet and BEANS benchmarks

Innovation

• Self-distillation with prototype-learning classifier
• Multi-taxa dataset for expanded training
• Source-prediction training criterion