Automatic detection of cow/calf vocalizations in free-stall barn

Precision livestock farming dictates the use of advanced technologies to understand, analyze, assess and finally optimize a farm’s production collectively as well as the contribution of each single animal. This work is part of a research project wishing to steer the dairy farms’ producers to more ethical rearing systems. To study cow’s welfare, we focus on reciprocal vocalizations including mother-offspring contact calls. We show the set-up of a suitable audio capturing system composed of automated recording units and propose an algorithm to automatically detect cow vocalizations in an indoor farm setting. More specifically, the algorithm has a two-level structure: a) first, the Hilbert follower is applied to segment the raw audio signals, and b) second the detected blocks of acoustic activity are refined via a classification scheme based on hidden Markov models. After thorough evaluation, we demonstrate excellent detection results in terms of false positives, false negatives and confusion matrix.