A method for automated individual, species and call type recognition in free-ranging animals

Abstract

The ability to identify individuals reliably is often a key prerequisite for animal behaviour studies in the wild. In primates, recognition of other group members can be based on individual differences in the voice, but these cues are typically too subtle for human observers. We applied a combined mechanism consisting of a call feature extraction (mel frequency cepstral coefficients) and pattern recognition algorithm (artificial neural networks) to investigate whether automated caller identification is possible in free-ranging primates. The mechanism was tested for its accuracy in recognizing species, call type and caller identity in a large population of free-ranging blue monkeys, Cercopithecus mitis stuhlmanni, in Budongo Forest, Uganda. Classification was highly accurate with 96% at the species, 98% at the call type and 73% at the caller level. It also outperformed conventional discriminant function analysis in the individual recognition task. We conclude that software based on this method will make a powerful tool for future animal behaviour research, as it allows for automatic, fast and objective classifications in different animal species.