Adaptive Loss Balancing for Multi-Task Bioacoustic Classification of Bird Species and Call Types

2026-07-03

AI Summary

This study addresses the challenges of label imbalance and multi-task loss coordination in joint classification of bird species and call types within passive acoustic monitoring. The authors propose a factorized multi-task classification framework and systematically investigate the interplay between domain-pretrained avian audio encoders—ConvNeXtBS, EAT, BirdMAE, and ProtoCLR—and various adaptation strategies (linear probing, attention probing, and full fine-tuning) combined with adaptive loss-balancing methods, including homoscedastic uncertainty weighting, dynamic weight averaging, and GradNorm. Experimental results demonstrate that lightweight adaptation with a frozen backbone achieves a superior trade-off between efficiency and performance. The factorized model substantially improves call-type recognition, with ConvNeXtBS yielding the best species classification under linear probing and BirdMAE excelling in call-type identification with attention probing. Adaptive loss weighting consistently enhances species recognition more reliably than call-type classification.

Abstract

Reliable analysis of bird vocalisations in passive acoustic monitoring requires models handling multiple, imbalanced annotation targets. We extend BirdCallNet for joint species and call-type classification on the long-tailed WiWa dataset and investigate how task-loss balancing interacts with pretrained representations and adaptation depth. We evaluate four bird-domain encoders, ConvNeXtBS, EAT, BirdMAE, and ProtoCLR, with separate species and call-type heads under linear probing, attentive probing, and full fine-tuning. A manually tuned fixed objective is compared with homoscedastic uncertainty weighting and Dynamic Weight Averaging across all three adaptation regimes, while GradNorm is evaluated only under full fine-tuning. Results indicate that the factorised multi-task formulation yields the most consistent improvements over the combined single-task baseline for call-type recognition, while its effect on species recognition depends on the adaptation regime. Full fine-tuning is not consistently optimal: ConvNeXtBS achieves the highest mean species performance under linear probing, whereas BirdMAE provides the strongest call-type performance under attentive probing. Adaptive weighting benefits species recognition more consistently than call-type recognition. Uncertainty weighting is particularly effective for species recognition under attentive probing, whereas Dynamic Weight Averaging is generally stronger for the same task under full fine-tuning. GradNorm achieves competitive call-type performance for selected backbones but consistently underperforms other weighting strategies for species recognition and incurs higher computational and memory costs. Overall, the preferred loss-balancing strategy depends on the backbone, adaptation regime, and target task, while frozen-backbone adaptation can provide a more favourable performance-efficiency trade-off than end-to-end fine-tuning.

Problem

• multi-task learning
• loss balancing
• bioacoustic classification
• class imbalance
• bird vocalisations

Innovation

• adaptive loss balancing
• multi-task bioacoustic classification
• pretrained audio encoders
• frozen-backbone adaptation
• long-tailed bird vocalization dataset