Big Data Approaches to Bovine Bioacoustics: A FAIR-Compliant Dataset and Scalable ML Framework for Precision Livestock Welfare

This record provides a FAIR-compliant bovine vocalization dataset and derived acoustic features used to develop and benchmark scalable machine-learning models for precision livestock welfare. The dataset contains 569 expertly curated cow vocalization clips spanning 48 behavioural classes, recorded across three commercial dairy farms in realistic barn environments (multiple barn zones and microphone setups). These clips were selected from approximately 90 hours of raw barn recordings and represent a range of welfare-relevant contexts, including estrus-related calls, distress and discomfort vocalizations, social communication, and feeding- or water-related anticipation calls. Each clip is accompanied by standardized metadata (e.g., anonymized farm ID, barn zone, microphone and placement context, recording date/day, and behavioural labels) and a set of 24 acoustic descriptors (duration, F0 statistics, intensity, formants, spectral descriptors, MFCC-based features, etc.) extracted using an open Python pipeline (Praat/Parselmouth, librosa, and related tools). An example pair of raw and denoised long recordings is included to illustrate the noise-reduction pipeline used prior to segmentation of individual calls. Files included in this record are: Dataset.zip – all final curated audio clips used in the study (one file per clip). metadata_full_clips.xlsx – per-clip metadata including recording context and behavioural labels. acoustic_features_24.xlsx – 24 acoustic features for each clip, aligned 1:1 with the metadata. raw_unfiltered_recording.WAV and denoised_recording.wav – an example long raw barn recording and its denoised counterpart. Data are shared under restricted access to protect farm confidentiality. Access requires a Data Access Agreement for non-commercial research use, including a commitment not to attempt re-identification of farms, animals, or people, and not to redistribute audio files outside approved projects. This dataset supports research on bovine bioacoustics, precision livestock farming, and animal welfare monitoring, and is associated with the manuscript “Big Data Approaches to Bovine Bioacoustics: A FAIR-Compliant Dataset and Scalable ML Framework for Precision Livestock Welfare.” Funding: The authors thank the Natural Sciences and Engineering Research Council of Canada (NSERC), Nova Scotia Department of Agriculture, Mitacs Canada, and the New Brunswick Department of Agriculture, Aquaculture and Fisheries for funding this study.