delphinID: deep learning models for automatic identification of Northeast Atlantic delphinid species from acoustic recordings of whistles and clicks

delphinID models are deep convolutional neural networks (CNNs) trained in Python Tensorflow 2.18.0 to automatically classify detections of delphinid whistles and clicks to species by latent patterns in average low-frequency power spectra calculated across groups, or frames, of detections. Separate models for classifying based on either whistle or click detections output predictions intended to be used as a feature vector for an event classifier, which predicts species based on information from both whistles and clicks. Cross-validated testing estimates delphinID to classify events with average accuracy ranging from 80% for Delphinus delphis to 92% for Lagenorhynchus albirostris. Species that delphinID can classify from acoustic recordings in the Northeast Atlantic are listed below:

• Atlantic white-sided dolphin (Lagenorhynchus acutus)
• Common bottlenose dolphin (Tursiops truncatus)
• Killer whale (Orcinus orca)
• Long-finned pilot whale (Globicephala melas)
• Risso's dolphin (Grampus griseus)
• Short-beaked common dolphin (Delphinus delphis)
• White-beaked dolphin (Lagenorhynchus albirostris)

Both click and whistle classifiers were trained and evaluated exclusively on visually-confirmed single species recordings of these species and only evaluated on recordings spatiotemporally independent from those used to train the models. delphinID is currently being used in various government-funded monitoring projects in Scotland and it is our hope that it will continue to facilitate acoustic monitoring of delphinids in northeast Atlantic waters in the future.

Use delphinID easily within the widely used open-source software PAMGuard. delphinID classifiers are stored in a compressed format compatible for use within PAMGuard's Deep Learning module. Using the configuration file provided, users can import the classifiers to classify whistle detections from the Whistle & Moan Detector module and click detections from the Click Detector module. The results of this base-level classification should be exported to a PAMGuard database file, which can then be used to classify events to species using the eventClassifier interface, which was written in R Shiny 1.9.1. This application integrates predictions from whistles and clicks to classify events and can be run to classify data either in near-real time or in post-processing. Please consult the README documents for guidance and tutorials on using delphinID models together with the eventClassifier application. PAMGuard can be installed here.