You may not know it, but fish can apparently actually be quite talkative. To analyze unique fish sounds, scientists spend hours listening to underwater acoustic recordings until they hear the sounds of whichever fish they’re looking for. It’s a slow and painstaking process, but you know what could make it a lot easier and faster? If you didn’t guess AI, you must be new here.
A team of scientists at the Woods Hole Oceanographic Institution has developed a new computer system that uses a neural network to listen to underwater recordings and pick out individual fish sounds. The system was trained on a database of known fish sounds and can locate the call of a specific species in a recording 25 times faster than a human.
It also works in real time, identifying the species that makes a specific sound as soon as it hears it. This means that if it were paired with a camera system, it could also theoretically pair previously unidentified sounds with species. The team is currently working on integrating the system with an autonomous underwater vehicle called CUREE, in the hopes that it could be sent out over coral reefs to identify both known and new fish sounds in real time.
