Simulation of Marine Mammal Calls in Deep-Sea Environment

In the deep-sea environment, the sound propagation is rich and complex, and the simulation of the original data of marine mammal calls is often not enough to meet the actual needs. Based on the deep-sea environment simulated by BELLHOP, we analyze the characteristics of different acoustic channels and build two types of generative adversarial network models, DCGAN and WGAN-GP, to simulate and generate marine mammal calls data with the characteristics of acoustic channels. The evaluation results show that the two generative adversarial networks models exhibit different simulation performance for the five marine mammal calls data, and both of them can effectively simulate samples with more channel characteristics and increase the variability between different channels.