An automatic acoustic bat identification system based on the audible spectrum

Nowadays the task of monitoring bat species is a very difficult task because of several factors. The main ones are the difficulty of creating databases automatically and the particularities of the vocalizations of bats. For this reason, it is common to extract bat calls manually from a recording and treat them individually. We propose a new form of identification and labeling process based on adapting bat calls to the audible spectrum and significantly reducing the noise of its spectrogram. This process can be performed automatically from a recording made in a natural area. Our database consists of 189 h of recordings obtained in various natural areas in Costa Rica. 50 bats calls of 7 different classes are extracted from this database. We have obtained an average error of 2.7% and 3 of the 7 classes have an error below 1%.