Benchmarking for the automated detection and classification of southern yellow-cheeked crested gibbon calls from passive acoustic monitoring data
Recent advances in deep and transfer learning have revolutionized our ability for the automated detection and classification of acoustic signals from long-term recordings. Here, we provide a benchmark for the automated detection of southern yellow-cheeked crested gibbon (Nomascus gabriellae) calls collected using autonomous recording units (ARUs) in Andoung Kraleung Village, Cambodia. We compared the performance of support vector machines (SVMs), a quasi-DenseNet architecture (Koogu), transfer learning with pretrained convolutional neural network (ResNet50) models trained on the ImageNet dataset, and transfer learning with embeddings from a global birdsong model (BirdNET) based on an EfficientNet architecture. We also investigated the impact of varying the number of training samples on the performance of these models. We found that BirdNET had superior performance with a smaller number of training samples, whereas Koogu and ResNet50 models only had acceptable performance with a larger number of training samples (>200 gibbon samples). Effective automated detection approaches are critical for monitoring endangered species, like gibbons. It is unclear how generalizable these results are for other signals, and future work on other vocal species will be informative. Code and data are publicly available for future benchmarking.