Acoustic classification of marine mammals by deep learning

Abstract

The classification of marine mammals has been an important topic for monitoring their population numbers. Many species of whales are considered endangered, making this task even more imperative. In this thesis two species of whales will be classified using deep learning with audio files as input data. A baseline model will be compared to the UPC?s Double Multi-Head Attention model. The baseline model consists of a CNN based model while the UPC system will help inspect the effect of attention in the task at hand. Furthermore, the effect of changing the Mel scale coefficient in the performance of the attention model will be investigated.