Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study introduces PULSE, a semi-supervised, multi-task framework designed for classifying Orthoptera bioacoustics, which addresses the limitations of existing automated tools that are often narrowly trained and non-transferable. By integrating weakly-supervised species classification, self-supervised learning on unlabelled audio, and knowledge distillation, PULSE significantly outperforms a state-of-the-art general model, achieving a macro F1 score of 0.21 and an AUC of 0.74, with further improvements through active learning. Additionally, the embeddings generated by the model reveal ecologically meaningful structures, facilitated by an interactive visualization tool for ecological discovery.
A semi-supervised framework for insect song classification not only outperforms existing models but also uncovers ecologically significant patterns in bioacoustic data.
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.