Abstract
With the advancement of precision livestock farming (PLF), acoustic technology has emerged as a key tool for tracking the health and well-being of laying hens, owing to its non-invasive, real-time and cost-effective nature. In this study, continuous audio data were collected from commercial chicken houses over a period of 15 days, in addition to temperature and humidity index (THI) analysis, to develop a convolutional neural network (CNN)-based model for classifying chicken squawks. This approach enabled the investigation of the relationship between environmental adaptability and acoustic traits in a mixed-sex rearing system. Significant daily variations were observed in the acoustic environment of the chicken house, with rooster crowing behavior corresponding to the highest noise levels (45-50 dB) recorded in the early morning hours. The CNN model achieved 98 % accuracy, along with both macro-average and micro-average scores of 98 %, in classifying roosters, hens, and other sounds, effectively addressing the issue of rooster crowing disturbances in mixed-rearing conditions. Additionally, the model revealed that fundamental frequency shift (F0 Shift) was positively correlated with normal egg production (r = 0.68, p = 0.025), while specific mel-frequency cepstral coefficients (MFCC_7, MFCC_10) associated with hen vocalization were significantly negatively correlated with THI ( r = -0.23, p < 0.05; r = -0.37, p < 0.001). These findings highlight the potential of acoustic monitoring as a novel dynamic method for evaluating environmental adaptability and health status in laying hens, reinforcing its utility in precision livestock farming under challenging rearing conditions.