Classification of stress calls of the domestic pig (Sus scrofa) using LPC-Analysis and a self organizing neuronal network

In the last years sound analysis has become an increasingly important tool to interpret the behaviour, the health condition, and the well-being of ani mals. The paper presents a procedure that allows to characterize, classify and visualize stress calls of domestic pigs (Sus scrofa) in three steps. (I) Starting from the acoustic model of sound production features are extra cted from the call using the linear prediction method. This procedure, line ar prediction coding (LPC), delivers an extremely compact short time repres entation of the call with a relatively low effort of calculation and a low number of features. (2) A neuronal network was trained such that topologica l relations of the neurons represent the input vector space of the determin ed LPC-coefficients. This resulted in a feature map, where the positions of the neurons allow conclusions about the structure of the input data. (3) V isualizations of the clustering structure of the calls were performed using various types of representations. The procedure now allows the development of online monitoring of stress calls in farming environments.