PeaDetect dataset: curated audio recordings of Indian peafowl (Pavo cristatus) vocalizations for bioacoustic monitoring

This data article describes PeaDetect, the first curated audio dataset specifically designed for detecting Indian peafowl (Pavo cristatus) vocalizations. The dataset contains 2,950 five-second audio clips (4.1 hours total duration), evenly balanced between peafowl presence (1,475 clips) and absence (1,475 clips) classes. Recordings were sourced from two public repositories: Xeno-Canto (332 original peafowl recordings from India and Sri Lanka, 2003-2024, other bird species vocalizations) and Freesound (environmental sounds and insects from Sri Lankan habitats). All audio was standardized to 44.1 kHz, 16-bit, stereo WAV format and segmented into 5-second clips. Each clip was manually validated by an ornithologist, with inter-annotator agreement assessed on 20% of the dataset (Cohen's kappa = 0.905). To prevent data leakage, source-aware 5-fold cross-validation splits are provided, ensuring all clips from the same original recording remain within the same fold. The dataset includes comprehensive metadata with geographic (87.1% India, 11.9% Sri Lanka, 1.0% other regions), temporal (2003-2024, all seasons), and acoustic specifications. This dataset supports research in bioacoustic monitoring, human-wildlife conflict mitigation, and machine learning for conservation.