Bird Call Identification

Abstract

The paper purposes an approach identify the bird species given their chirping sounds. The paper proposes Convolutional Neural Network architecture for classifying the species. In our proposed approach, we first convert the audio files to Mel spectrogram and use them for classification. We built a CNN model with three Convolutional layers, and then tuned parameters like number of epochs, learning rate, dropout percentage and size of convolutional matrix. The training dataset consists of 1485 audio clips containing 40 distinct species found across South Africa. The designed set of model achieved accuracy in the range of 32–52% on the test dataset.