Understanding livestock’s emotional states (affective states) is crucial, and vocalizations offer a promising avenue for non-invasive assessment. While research has explored vocal indicators in pigs, horses, poultry, and goats, cattle remain understudied. Using sensors to collect biometric data as a means of measuring animal emotions is a topic of growing interest in agricultural technology. Cows produce distinct low and high-frequency vocalizations associated with close-range and long-distance communication, respectively, with the latter potentially linked to negative emotions. Additionally, their vocalizations are shown to be unique to each individual across various contexts. Given the challenges dairy cows face during distress, pain, or fear, studying their vocalizations during negative states is particularly important. Artificial intelligence can help achieve this objective along with anomaly detection and continuous monitoring. Further, continuous cattle monitoring requires low latency for quick actions, low cost, and scalability. Therefore, this study proposes a solution by porting a small CNN-LSTM model on a low-cost Jetson Nano edge device that can identify animal affective states for two high-frequency and low-frequency vocalization classes using 3D MFCC features in dairy cows using the cattle vocalization dataset. The training and testing accuracy of a model is 96.92% and 95.81% respectively is higher than state-of-the-art methodologies.