Classification of Bird Species using Audio processing and Deep Neural Network

Abstract

Ornithology deals with the methodological study of birds and it consists of knowledge of birds and all other things which relates to them. The people who work in this field are knowns as Birders. One of their main tasks is to find and track some rare species of birds. Classification of a bird species can be a very challenging task and especially based on the sound they produce. Classification of bird species based on audio can be very helpful to find some rare birds as it will tell the species of the bird before actually tracking it and this can save a lot of time and efforts. In order to reduce this time ands efforts a lot of researchers have proposed different methods and techniques like machine learning, deep learning and wavelet study to classify the species of a bird. Some of these techniques are VGG16 and ResnetV2, Multilayer Perceptron, Naïve Bayes, and J4.8 Decision tree, Convolution Neural Network and many more. Along with this they have used Audio feature extractors like spectrogram, Inverse Short Time Fourier Transform and Mel-frequency cepstral coefficient. These techniques are discussed in detail in the related work section of this paper. In order to classify the species of a bird in this paper we have used several preprocessing techniques like conversion of .mp3 audio file to .wav audio files, feature extraction, class balancing also we have used one approach to minimize the time to pre process these audio files and we have also used to the Mel-frequency cepstral coefficient to extract features from these audio files and finally we have used a Deep Learning based Classification model to classify and predict the species of a Bird by analyzing the sound produced by that bird.