Application of time domain signal coding and artificial neural networks to passive acoustical identification of animals

The development of a computer-based system capable of identifying animals automatically from sounds they generate is described. The system uses time domain signal coding techniques and artificial neural networks for discriminating between the sounds of different animal species within a group. Results for British species of Orthoptera (bush crickets and crickets) show that it is capable of discriminating 13 species with 100% success and zero misidentification under low noise conditions. Results are also given for noise tests using 25 species of Orthoptera (including grasshoppers). The approach can be applied to other animal groups such as birds, mammals and amphibia; preliminary results are also presented for tests with 10 species of Japanese bird. The approach is generic and has application in many fields including non-destructive testing and physiological signal analysis. (C) 2001 Elsevier Science Ltd. All rights reserved.