UVic Biologists Harness AI to Decode Unique Fish Sounds

Biologists from the University of Victoria have made a significant breakthrough in understanding fish communication by using artificial intelligence (AI) to identify and differentiate the sounds made by various fish species. This research reveals that even closely related species produce unique and recognizable sounds, allowing researchers to distinguish between them with impressive accuracy.

Utilizing passive acoustics, the team identified distinct sounds for eight fish species found around Vancouver Island. They developed a machine learning model capable of predicting which sounds correspond to which species with an accuracy rate of 88 percent. According to Darienne Lancaster, a PhD student at UVic and lead researcher on the project, “We knew previously that many fish were making sounds in the wild, but we didn’t know which sounds belonged to which species, or if it was possible to tell these sounds apart.”

One notable example is the black rockfish, which emits a long, growling sound reminiscent of a frog’s croak. In contrast, the quillback rockfish produces a series of short knocks and grunts. Lancaster expressed her excitement about the findings, stating, “It has been exciting to see how many different species of fish make sounds and the behaviours that go along with these calls.” She noted that some fish, like the quillback rockfish, produce rapid grunting sounds when threatened, suggesting these sounds may serve as a defensive mechanism. Other species, such as the copper rockfish, make knocking sounds while hunting prey along the ocean floor.

To gather the data for this research, Lancaster employed a technique known as passive acoustic monitoring. This involved collecting underwater audio and video using a sound localization array designed by Xavier Mouy, a former UVic PhD student and project collaborator. By analyzing sound characteristics, the researchers were able to identify differences in the calls of various species.

The machine learning model developed by Lancaster examined a set of 47 different sound features, including duration and frequency, to detect subtle distinctions in the sounds produced by each species. This analysis enabled the model to effectively group species calls together based on these small variations.

The innovative techniques developed by the research team have the potential to be adapted by scientists worldwide, facilitating the study of other fish calls. This research was funded by the Natural Sciences and Engineering Research Council of Canada and Fisheries and Oceans Canada, highlighting its relevance and importance in the field of marine biology.

As researchers continue to explore the acoustic world of fish, this study opens new avenues for understanding the complex behaviors and communications of aquatic life, providing valuable insights into the ecosystems they inhabit.