Field monitoring of bat species by means of acoustic methods has been the start of the new epoch in the history of bat research. Our aim was to investigate these methods are adaptable to shrew species. We have made sound recordings in the audible range from 6 shrew species in standard environment. In addition we trapped animals and made long field recordings in different natural habitats in Hungary. We applied artificial intelligence methods in Matlab environment for solving different informatic problems. First problem was to recognize automatically shrew calls in the long noisy field recordings. After manual selection of shrew calls we have trained an Artificial Neural Network to collect automatically these calls from recordings. Comparing the acoustic and trapping data it turned out that this acoustic method shows shrew activity in much finer resolution. The second problem was the species identification. Applying Hidden Markov Models we reached at least 70% correct identification. We recognized 2 or 3 different group of calls from different individuals of the same species, which may show us more information about the individuals. We think these automatic methods may serve the better understanding of ethology and ecology of shrews, which may lead making more effective conservation program.
Abstract: