Bangwa Warbler Acoustic Recogniser Data

This dataset contains WAV recordings and PNG spectrograms of the African songbird Bangwa Warbler (Bradypterus bangwaensis). The PNG spectrograms were used to train and evaluate an automated ResNet18-based classifier, while the WAV recordings were used to create a custom BirdNET species‑specific model (BirdNET-Analyzer version). The dataset is organised into folders representing training, validation, and testing sets. This dataset accompanies the publication: Winiarska, D., Budka, M. Reproducing a year‑round expert survey with deep learning. Code for using the PNG dataset, including the full training and evaluation pipeline, is available at: https://github.com/dwiniarska/Bangwa-Warbler-Acoustic-Recogniser Folder structure: PNG │ ├── training_thresholds/ # PNG sets used for training with reduced data │ ├── train/ # Training spectrograms (default 320 positive/960 negative) │ ├── val/ # Validation spectrograms │ └── test/ # Test set spectrograms WAV │ ├── training_samples/ # 5-sec audio clips used for training │ ├── validation/ # 5-sec audio clips used for validation │ └── test/ # 1-minute audio files for year-round inference