Neural Network Identifies Coral Reef Sounds

Despite making up less than 1% of the world’s oceans, one-quarter of all marine species spend some portion of their life on a coral reef.

With so much life in one spot, researchers can struggle to gain a clear understanding of which species are present and in what numbers.

In JASA, published on behalf of the Acoustical Society of America by AIP Publishing, researchers from Woods Hole Oceanographic Institution (WHOI) combined acoustic monitoring with a neural network to identify fish activity on coral reefs by sound.

For years, researchers have used passive acoustic monitoring to track coral reef activity. Typically, an acoustic recorder would be deployed underwater, where it would spend months recording audio from a reef. Existing signal processing tools can be used to analyze large batches of acoustic data at a time, but they cannot be used to find specific sounds. To do that, scientists usually need to go through all that data by hand.

As an alternative, the researchers trained a neural network to sort through the deluge of acoustic data automatically, analyzing audio recordings in real-time. Their algorithm can match the accuracy of human experts in deciphering acoustical trends on a reef, but it can do so