In recent years, sound analysis has become an increasingly important tool for interpreting animal behavior, health condition and wellbeing. Additionally, vocalisation may provide an objective, non-invasive and useful tool for evaluating the emotional state of animals under natural or captive conditions.

A sound analysis procedure based on linear prediction coding (LPC) and a self-organising neural network is presented, and its capacities are demonstrated using various stress calls of domestic pigs. Using LPC, an extremely compact, short-time representation of the call was obtained, comprising only a few features (LPC-coefficients). A neural network was trained such that topological relations of the neurons represent the input vector space of the determined LPC-coefficients. Hence, the resulting feature map allows conclusions to be drawn concerning the structure of the input data.

Early results demonstrate that the procedure is able to distinguish stress calls from any other calls or noise. It is also possible to discriminate individuals on the basis of their calls, and to differentiate between stress-related calls caused by a variety of sources (e.g. normal handling versus castration of young piglets).

This procedure may be used as a methodological approach to solve different analysis and classification tasks in animal vocalisation. Under the assumption that the observed animals vocalise when stressed, the procedure can be executed in pseudo real-time on commercial laptops, to allow automatic monitoring of the vocal behavior of pigs in various stressful situations (e.g. housing and transport).

Paper presented at Measuring Behavior 2002 , 4^th International Conference on Methods and Techniques in Behavioral Research, 27-30 August 2002, Amsterdam, The Netherlands

© 2002 Noldus Information Technology bv