English

Abstract

Acoustic recordings of birds have been used by conservationists and ecologists to determine the population density and bio- diversity of bird species in a region. However, it is hard to analyze and visualize the presence/absence of a specific bird species by aurally hearing these recordings even by an expert bird song specialist. In this paper, we present a computational tool to cluster and recognize bird species based on their sounds and visualize relationships of within-species and between-species sounds based on their similarity measures. The tool has been evaluated on two datasets of varying complexity containing acoustic recordings of eleven birds’ songs and calls using various similarity measures. Principal Component Analysis (PCA) was used for feature selection. Euclidean distance, Mahalanobis distance, and cosine similarity among features was used for pair-wise similarity calculation. The results of similarity measures have been compared using 3-fold cross-validation and validated by spectrograms patterns obtained from frequency representation of acoustic recordings of the selected birds’ songs and calls. Cosine similarity performed better to measure underlying patterns of birds’ sounds and identify mutual relationship among species. It was concluded that the proposed tool can be used as a novel method for conversationalists, ecologists, ornithologists, and evolutionary scientists as well as tourists and bird watchers to recognize different birds’ species, study their mutual relationship, locate the area with highest population density, estimating the predators, and biodiversity in a specific region.