Acoustic Identification of Nocturnal Bird Species

Abstract

Automatic classification of bird species based on their vocalizations is a topic of crucial relevance for the research conducted by biologists, ornithologists, ecologists, and related disciplines. This work in concentrated on nocturnal species; even though the analysis of their population trends is a key indicator, there is a gap in the literature addressing their audio-based identification. After compiling a suitable dataset including six nocturnal bird species, this study employs both supervised (k-Nearest Neighbors, Support Vector Machines) and unsupervised (k-means) methods operating in the feature space formed by Mel-spectrograms. We conclude that automatic classification based on k-Nearest Neighbour and Support Vector Machines provide almost excellent results with a recognition rate in the order of 90%.