Noise robust bird call localisation using the generalised cross-correlation with phase transform in the wavelet domain.

Abstract

Bird localisation using passive acoustic methods is a non-intrusive solution for taking a census of bird species. Through the recordings from a microphone array, the generalised cross-correlation with phase transform (GCC-PHAT) has been adopted widely in the estimation of the direction-of-arrival (DOA) of audio signals especially speech in indoor environments, as it performs very well in a reverberant environment. However, the performance is degraded when the signal to noise ratio is low. This study investigates the performance of DOA estimation when the GCC-PHAT is applied in the wavelet domain. Three configurations are considered in this study and the performance of these configurations is assessed by numerical simulation under ideal setup and practical experiments using audio signals recorded in a typical native forest in New Zealand. The results suggest the configuration which applies GCC-PHAT and denoising in the wavelet domain and estimates DOA from the reconstructed cross-correlation can give more accurate estimation compared to the conventional GCC-PHAT.