Abstract: In this research, a system for accurately identifying bird species was developed, and tactics for identifying them were studied. Automatic recognizing bird sounds with no physical interaction has proven to be a challenging and time-consuming task for significant research in ornithology's taxonomy and other subfields. Birds make a wide range of vocalisations, and different species of birds have distinct functions.
Manual annotation of each recording is used in current methods for processing big bioacoustic datasets. This necessitates specialised expertise and an inordinate amount of time. Recent advances in machine learning had made it easier to identify specific bird sounds for popular species with enough training data. However, developing such tools for rare and endangered species remains difficult.
This problem has been addressed in two stages : pre-processing and modelling (CNN model). The first step is to create a spectrogram from an audio input. The spectrograms were used as input in the second stage, which required establishing a neural network. Depending on the input properties, the Convolutional Neural Network identifies the sounds clip as well as separates the bird species.
Key-words: Bird Species Classification , Bird audio , method of pre-processing of bird sound, Convolutional Neural Network (CNN), Spectrogram
|
DOI:
10.17148/IJARCCE.2022.11629
