AUTOMATED BIRD DETECTION IN AUDIO RECORDINGS BY A SIGNAL PROCESSING PERSPECTIVE

In this study, an effective automated technique for the detection of bird sounds is presented in a signal processing perspective. The detection of bird sound by examining the sound patterns is the basic step for wildlife monitoring. An Automated Bird Detection (ABD) system based on Dual-tree M-band Wavelet transform (DMWT) is designed. The more intrinsic content of the audio is extracted as features by DMWT and this is the crucial stage as the extracted features directly influence the efficiency of the ABD system. It classifies the given audio signals into two classes; birds are present or not present. The sounds in the audio signals are modeled by Gaussian Mixture Model (GMM) with the help of DMWT features. The ABD system is analyzed by changing the DMWT decomposition level, and Gaussian components used to model each fault. Results show that the ABD system achieves 97.82% accuracy by 3rd level DMWT features when modeled by 16 Gaussian components.