Wavelet-Based Acoustic Classification of Bird Species

Abstract

Identifying birds based off their calls is extremely useful in the realm of avian biology, especially ecology. In this paper we consider the calls of the Whip-poor-will (Antrostomus vociferus), the Northern Bobwhite (Colinus virginianus), the Barred Owl (Strix varia), the Eastern Kingbird (Tyrannus tyrannus), and the Common Raven (Corvus corax), and ways of using automated classifiers to identify the bird species based off these calls. In this study, we segment the bird calls into syllables. Then we apply wavelet decomposition to decompose the recordings of the bird calls and extract certain parameters from the syllables. All of the instances and the parameters were placed in an Excel file and uploaded into WEKA, a software for classification. We used various classifiers to classify the different syllables, but Random Tree and Random Forest were the most successful in our study. Both of the classifiers achieved over 70% accuracy when classifying species on the data set that contained the various species of birds. This thesis shows that birds can be classified into their species based off recordings of their calls with relative accuracy.