Decoding Insect Song: A Multitask Semisupervised Orthoptera Bioacoustic Classifier

2026-06-11

AI Summary

Existing automated acoustic monitoring tools struggle to effectively identify orthopteran insect calls due to narrow training data and poor generalization. This work proposes PULSE, a novel framework that uniquely integrates multi-task semi-supervised learning with knowledge distillation, combining weakly supervised species classification, self-supervised pretraining on unlabeled field audio, and knowledge transfer from a general bioacoustic model. Furthermore, active learning is incorporated to optimize annotation efficiency. The proposed method substantially outperforms generic models, achieving an F1 score of 0.21 versus 0.07 and an AUC of 0.74 versus 0.45. When combined with active learning, performance further improves to F1 = 0.34 and AUC = 0.84. Embedding visualizations also demonstrate the framework’s capacity to support ecological discovery.

Abstract

Passive acoustic monitoring holds great promise for ecological inference, yet existing automated tools are typically narrowly trained and non-transferable. We address these limitations with PULSE, a semi-supervised, multi-task framework for Orthoptera bioacoustics, combining weakly-supervised species classification, self-supervised learning on unlabelled field audio, and knowledge distillation from a general-purpose bioacoustic model. Our domain-adapted specialist model outperforms a state-of-the-art general model across all metrics (macro F1: 0.21 vs. 0.07; AUC: 0.74 vs. 0.45; AP: 0.32 vs. 0.19), with active learning further raising F1 to 0.34 and AUC to 0.84. Beyond classification, the learned embeddings encode ecologically meaningful structure, exposed through an interactive visualisation tool for ecological discovery.

Problem

• bioacoustic classification
• Orthoptera
• passive acoustic monitoring
• transferability
• insect song

Innovation

• semi-supervised learning
• multi-task learning
• bioacoustic classification
• knowledge distillation
• active learning