Machine learning for acoustic source localization in room and ocean acoustics

Despite decades of development, acoustic source localization has been challenged by mislocations. Instead, we learn the localizations directly from data using machine learning. After a mapping has been learned we test that the mapping generalizes well on test data. We demonstrate the approach in an ocean waveguide as a classification and a regression neural network, in a room acoustic setting we demonstrate how a neural network can alleviate multipath.