Animal Sound Classification using Sequential Classifiers

Abstract

Several authors have shown that the sounds of anurans can be used as an indicator of climate change. But the recording, storage and further processing of a huge number of anuran’s sounds, distributed in time and space, are required to obtain this indicator. It is therefore highly desirable to have algorithms and tools for the automatic classification of the different classes of sounds. In this paper five different classification methods are proposed, all of them based on the data mining domain, which try to take advantage of the sound sequential behaviour. Its definition and comparison is undertaken using several approaches. The sequential classifiers have revealed that they can obtain a better performance than their non-sequential counterpart. The sliding window with an underlying decision tree has reached the best results in our tests, even overwhelming the Hidden Markov Models usually employed in similar applications. A quite remarkable overall classification performance has been obtained, a result even more relevant considering the low quality of the analysed sounds.