Author: Klaus Riede

Affiliation: Stuttgart State Museum of Natural History

Title: Computer-aided inventories of bioacoustic diversity: first results and future needs

Session:

# Monitoring the Environment: Bioacoustics as a tool for inventory and monitoring of biodiversity

Abstract

The reliable detection and registration of animal calls allows assessment and monitoring of many vertebrate and insect species. Non-invasive acoustic registration is particularly useful in aquatic habitats or forests, where visual observations are difficult or even impossible. However, bioacoustic identification is time-consuming and requires expert knowledge. Therefore, computer-aided classification tools of animal calls have been developed for a wide variety of groups reaching from insects to dolphins, using distinct techniques and software environments. However, none of these approaches developed into a well-documented standard, and the underlying collections of recordings are often not available for further research. To illustrate future needs, a suite of neural networks and tools for acoustic classification is presented, based on the DORSA (Digital Orthoptera Specimen Access - www.dorsa.de) sound repository. DORSA provides online access to a collection of approximately 5,000 digitised Orthoptera songs with reference to voucher specimens. A small subset, representing only 40 ensiferan species (Grylloidea and Tettigonioidea) and 200 recordings, was used to program and train neural networks for species identification, with an accuracy of more than 90% (Dietrich et al. 2004, Pattern Recognition 37, 2293). In addition, feature extraction tools were used to label all DORSA cricket songs (Grylloidea) with their respective carrier frequency and pulse distance, serving as look-up tables to compare stored songs with new registrations. In spite of these promising results, an efficient connection and data flow between neural network tools, DORSA database and users has not yet been established. This would require distinct tools such as web-based user interfaces and applications running on portable computers, to allow classification and identification in the field. The requirements for such tools are outlined, and the bioacoustic community is asked to share resources towards their development, particularly because they are urgently needed in endangered habitats such as rainforests.