Acoustic Monitoring of Singing Insects

Abstract

This work reports recent progress towards the development of a pilot system for automatic identification of singing insects. We propose a sound parameterization technique that is designed explicitly for the needs of acoustic insect recognition. It is combined with state-of-the-art classification methods that dominate speaker recognition technology. Specifically, the categorization of acoustic emissions of insects takes place on the levels of suborder, family, subfamily, genus and species. We evaluate the performance of our approach on a large and well-documented catalogue of recordings of crickets, katydids and cicadas. Identification accuracy that exceeds 98 % on the levels of suborder and family, and 86 % on the level of specific species out of 313 species is reported.