Dataset for: Automated detection of gibbon calls from passive acoustic monitoring data using convolutional neural networks in the "torch for R" ecosystem.

This is the supporting data for: Clink, D.J., Kim, J., Cross‐Jaya, H., Ahmad, A.H., Hong, M., Sala, R., Birot, H., Agger, C., Vu, T.T., Thi, H.N. and Chi, T.N., 2025. Automated detection of gibbon calls from passive acoustic monitoring data using convolutional neural networks in the “torch for R” ecosystem. Ecology and Evolution, 15(7), p.e71678. Summary of acoustic and image data. Data are in the "data_only" folder. Please see publication for details: https://doi.org/10.1002/ece3.71678. Dataset Dataset ID Country Species Number of Gibbon samples Number of Noise samples Sample rate Danum Valley Conservation Area Training Data Train-Val Malaysia Grey gibbon 502 2,254 16 kHz Jahoo Training Data Train-Val Cambodia Crested gibbon 213 2,130 32 kHz Maliau Basin Test Data Test-Generalizable Malaysia Grey gibbon 147 81 48 kHz Dakrong Nature Reserve Test Data Test-Generalizable Vietnam Buff cheeked gibbon 45 173 16 kHz Danum Valley Test Data Test-Final Malaysia Grey gibbon 383 2,905 16 kHz Jahoo Test Data Test-Final Cambodia Crested gibbon 296 5,684 32 kHz Summary of data and scripts. To run the scripts download the "data" and "results" folders from Zenodo and add to your project directory. Link to GitHub Repository: https://github.com/DenaJGibbon/torch-for-R-gibbons R Script Summary of Referenced Folders Top-Level Data or Results Folders Referenced Part 1a. Variability benchmarking results Training/test image folders (Danum, Jahoo, Combined); model run results for variability benchmarking. data/training_images_sorted/, results/part1/ Part 1b. Evaluate variability benchmarking results Evaluation output folders. results/part1/ Part 2a. Train CNNs over multiple epochs Training/test data for all sites; performance outputs for initial CNN evaluations (Cambodia, Malaysia). data/training_images_sorted/, results/part2/ Part 2b. Train CNNs (cont.) Model output folders (binary & multi-class); rerun AUC updates; model type labels. results/part2/ Part 3a. Data augmentation Augmented training images (Danum, Jahoo, Combined); test sets; model run outputs based on augmented data. data/DataAugmentation/, data/training_images_sorted/, results/part3/ Part 3b. Evaluation data augmentation Outputs from evaluating models trained on augmented data; model labels by type. results/part3/ Part 3c. Eval. augmentation on different test set Evaluation outputs for external test data (Maliau + Vietnam); multi-class model run outputs. TestData/, results/part3/ Part 4b. Comparison with BirdNET BirdNET outputs for Jahoo, Danum, and Combined data; test output from jittered data augmentation. results/part4/, results/part3/ Part 5. Final model performance Final evaluation images (per site); top-performing model folders; final performance CSVs. data/CombinedImagesWAEvaluation/, results/part5/ Part 6. Deploy Model over PAM data Longer sound files (Jahoo); final trained model checkpoint (.pt file) used for inference. data/WideArrayEvaluation/, results/part3/ Part 7. Call Density Plots Files for call density visualizations: GPS, selections, manually verified wavs or images (TP/FP). data/calldensityplots/