Expert-annotated whistle-type detection dataset for OpenWhistle. Each example is a fixed-length 0.5 s audio window labeled with the whistle types present in that window; background/no-whistle windows are represented by an all-zero target vector.
Task: multi-label whistle-type detection on fixed-length audio windows
Target vector: label, with one binary decision per whistle type in the order SW_Neo, SW_Luna, SW_Nikita, SW_Nana… See the full description on the dataset page:
https://huggingface.co/datasets/OpenWhistleNeurIPS26/OpenWhistle-Detection-Finetuning.
