Scientists discover how to 'speak like a fish' by matching sounds to species

Researchers off Vancouver Island matched over 1,000 underwater sounds to eight rocky reef fish species using an acoustic localization array and video, and a machine-learning model identified species from sounds with up to 88% accuracy.

Researchers recorded and matched underwater noises to specific fish species off the coast of Vancouver Island. The team combined an acoustic localization array with underwater video to pinpoint which animals produced each sound. This approach addresses a long-standing difficulty in marine study because sound travels far and rapidly underwater. The research was carried out in Barkley Sound and focused on rocky reef species. Key findings: - More than 1,000 fish sounds were documented and linked to eight species: copper, quillback, black, canary and vermilion rockfish, lingcod, pile perch and kelp greenling. - Canary and vermilion rockfish had not previously been documented producing sounds. - A machine-learning model trained on 47 sound characteristics could identify species from sound with up to 88% accuracy. - Some species produced distinctive calls: black rockfish made long, low growling sounds, while quillback rockfish produced short knocking and grunting noises. - Sounds were recorded during a wider range of behaviours than previously noted, including courtship, territorial displays, fleeing from predators, feeding and aggressive interactions. - Stereo cameras showed smaller fish generally produced higher-pitched sounds, suggesting size may relate to sound frequency. Summary: The study demonstrates a method for linking wild fish sounds to identified species by combining acoustic localization, video, and machine learning. The researchers plan to develop species-specific acoustic detectors and to investigate whether fish sounds vary by region, which would inform how passive acoustic monitoring might be used in future fisheries and conservation work.