Artificial neural networks: A new tool for studying lemur vocal communication

Abstract

Previous studies have applied Arti ficial Neural Networks (ANNs) successfully to bioacoustic problems at different levels of analysis (individual and species identi fication, vocal repertoire categorization, and analysis of sound structure) but not to nonhuman primates. Here, we report the results of applying this tool to two important problems in primate vocal communication. First, we apply a supervised ANN to classify 222 long grunt vocalizations emitted by five species of the genus Eulemur. Second, we use an unsupervised self-organizing network to identify discrete categories within the vocal repertoire of black lemurs ( Eulemur macaco ). Calls were characterized by both spectral (fundamental frequency and formants) and temporal features. The result show not only that ANNs are effective for studying primate vocalizations but also that this tool can increase the ef ficiency, objectivity, and biological signi ficance of vocal classi fication greatly. The advantages of ANNs over more commonly used statistical techniques and different applications for supervised and unsupervised ANNs are discussed.