Abstract
Rodent pups emit ultrasonic vocalizations (USVs) to solicit maternal care, enhancing survival. Inflammatory stressors, such as maternal high-fat diet (mHFD) and neonatal lipopolysaccharide (nLPS) exposure, affect neonatal neurodevelopment and behavior. Although basic features of USVs are widely used to assess these impacts, their sonographic and syntax characteristics remain underexplored. We employed DeepSqueak, a deep learning system, for automated unsupervised USV classification and detailed analysis of temporal and syntax changes, comparing it to manual coding for call number, duration, and frequency. Offspring of Long Evans rat dams, fed a high-fat or control diet for three weeks premating and throughout gestation and lactation, received 0.05 mg/kg intraperitoneal nLPS or saline on postnatal days (PND) 3 and 5 (n = 5-6 per condition, total N = 41). On PND 7, USVs were recorded after a brief maternal separation. DeepSqueak yielded 97% concordance (p > 0.8 vs. manual); mHFD increased bout duration by 25% (p < 0.01), and nLPS increased calls per sequence by 30% (p < 0.05). DeepSqueak matched manual accuracy, with automated analysis revealing that mHFD and nLPS significantly prolonged sequence durations and altered syntax transition probabilities. These findings suggest that temporal and syntactic USV features are sensitive markers of early inflammatory stress.