Three-dimensional acoustic-based model for abnormal cough sound localization: A proof-of-concept study in laying hens

With the rapid expansion of large-scale poultry farming, timely detection of abnormal laying hens is critical for early disease prevention. This study presents a three-dimensional (3D) sound source localization method based on the time difference of arrival (TDOA) and the steered response power–phase transform (SRP–PHAT) algorithms, enhanced by parabolic interpolation and particle swarm optimization. These enhancements address limitations of traditional methods, including low delay estimation accuracy and time-consuming global searches. Laboratory experiments utilized 520 audio samples collected from 26 spatial locations. Results indicated that the proposed method achieved centimeter-level spatial resolution and second-level response time. The mean absolute errors (MAE) for azimuth and elevation were 1.9° and 2.8°, respectively, with a distance MAE of 0.24 m. Errors in the X, Y, and Z directions were all within 0.15 m. To further evaluate system performance under practical environmental conditions, field experiments were conducted at 15 typical locations in a laying hen house using loudspeaker playback pre-recorded cough sounds to validate system performance under practical acoustic conditions. Despite multipath reflections and complex background noise, the system maintained strong localization performance during playback-based field validation, with mean absolute errors of 2.73° in azimuth, 3.18° in elevation, and 0.16 m in distance—closely aligning with laboratory results. Overall, the proposed method demonstrates favorable localization accuracy, robustness, and adaptability in both laboratory and field environments. This study provides a technical proof-of-concept for three-dimensional localization of replayed abnormal coughing sounds signals for laying hens.