Bird Sound Classification Using Deep Learning

Abstract

With the high advancements in technology and cellular world, comes the frowning situation of endangering bird species. Bird sound classification and prediction model is a step towards the process of conserving the bird species. The deep learning advancements helps to bring the solution to the problem of bird species identification from audio recordings. It is efficient and economical to identify and classify different bird species by audio recordings using machine learning algorithms and artificial models. Of course, there are challenges like choosing the right model that would minimize the signal to noise ratio between the training and test datasets. Observing different experiments on a few models, CNN( Convolutional Neural Network ) and ANN( Artificial Neural Network ) promised highest accuracy of which ANN showed results. During the analysis of the dataset and model, the implementation graphical representation of the model performance and visual understanding of the dataset enhance the observation and provide more insightful knowledge of the dataset. Bird identification is a difficult undertaking that frequently results in ambiguous labeling. Even skilled bird watchers dispute the species of a bird when presented with an image of one. It’s a demanding task that tests both people and computer’s visual talents. Although different bird species have the same core set of elements, their shape and appearance can change drastically. Due to differences in lighting and background, as well as great diversity in stance, intraclass variance is high (e.g., flying birds, swimming birds, and perched birds that are partially occluded by branches). Human knowledge of a species is insufficient to reliably identify a bird species, as it necessitates a great deal of skill in the subject of ornithology. Our study intends to use machine learning to assist amateur bird watchers in identifying bird species from photographs This research provides a deep neural network-based automate model for automatically identifying the species of a bird supplied as the test dataset. For 400 different bird species, the model was trained and evaluated.