This research delves into the automated classification of marine mammal species based on their intricate acoustic signals, employing Convolutional Neural Networks (CNN). As marine mammals rely extensively on acoustic communication, deciphering their vocalizations holds the key to understanding their behaviour and conserving these fascinating creatures. The study utilizes the Watkins Marine Mammal Sound Database, a rich collection of annotated underwater recordings, to train and evaluate the CNN model. The architecture involves convolutional layers for hierarchical feature extraction from spectrogram images, ultimately leading to multi-class classification. Results demonstrate the CNN’s commendable performance, with a Weighted F1 Score of 0.79 and an overall accuracy of 0.87. These metrics highlight the model’s robustness in handling class imbalances and its efficacy in accurately identifying diverse marine mammal sounds. The integration of CNNs in marine mammal research signifies a paradigm shift, providing a more efficient and scalable approach to studying and conserving these extraordinary creatures. The outcomes promise not only to unravel the mysteries of marine mammal acoustic communication but also to contribute valuable insights for conservation efforts in our interconnected marine ecosystems.