Seasoned birdwatchers can often identify species by their calls. Imagine doing the same for fish.

New research from the University of Victoria shows that closely related fish species produce distinctive sounds that can be separated by AI. This opens the door to monitoring programs that rely on acoustic signals rather than visual sightings.

The study, led by PhD student Darienne Lancaster, appeared in the Journal of Fish Biology.

Fish vocalizations have long been known, but distinguishing them in the wild was a challenge. As Lancaster explained, “We didn’t know which sounds belonged to which species, or if it was even possible to differentiate them.”

Using passive acoustic monitoring, the team collected underwater recordings while simultaneously verifying species with visual data.

They then trained a machine‑learning model that correctly matched sounds to species with 88 % accuracy, identifying eight British‑Columbia fish species.

The dataset also revealed behavioral insights, such as the quillback rockfish’s rapid grunting when pursued—likely a defensive response.

“It’s exciting to see how many species use sound and the contexts in which they do so,” Lancaster said.

Beyond basic science, the model offers a less invasive way to study fish behavior and monitor populations.

Future work will expand the database to include more species and explore regional variations within species, underscoring the vast diversity of marine life.