| Resumo: | In this dissertation we approach the problem of performing bioacoustic classification of four different small dolphin species by using their vocalizations. Cetaceans, (the taxonomic order which dolphins are part of) live in complex social societies and have been known to possess remarkable cognitive skills, being praised to have great intelligence capabilities. Cetaceans are most well known for their intricate communication patterns, which serve different purposes from mating advertisement to individual recognition. The analysis of these vocalizations, due to their intricacy has been for decades a laborious manual task, which takes a long time for specialists to perform. Our interest is in aiding researchers by developing machine learning methods capable of the analysis and classification of great volumes of cetacean recordings. We propose a four stage method which is capable of extracting relevant features from dolphin vocalizations making it possible to identify the corresponding species with great accuracy (achieving model accuracies above 95%). Although the resulting model is tai- lored to the classification of cetacean species indigenous to the Madeira Archipelago, which is expected to help the Madeira Whale Museum’s conservation efforts of these animals, it can be the foundation for future classifications of other cetacean species.
|
|---|