A near real-time orca detector using sound and acoustic context awareness is being developed for the Salish Sea, with potential uses including the identification of these mammals for vessels in the area.
Fabio Frazao, one of the first recipients of the $20,000 Roy Hyndman Ocean Observing Award, has completed his one-year project with Ocean Networks Canada (ONC). Using hydrophone data from ONC’s Victoria Experimental Network Under the Sea (VENUS) observatory, Frazao expanded upon his artificial intelligence (AI) algorithm to detect and classify orca sounds in the Salish Sea.
Orcas make a wide range of sounds like whistles, echolocation clicks, and low-frequency pops, which can be mistaken for other marine mammals and masked by ambient underwater noises.
The AI algorithm learned from thousands of these orca vocalizations and other sounds recorded from ONC’s hydrophone network deployed in the Salish Sea. The detector was then tested on 65 hours of acoustic data to detect and distinguish orca sounds.
A hydrophone array being prepared for deployment in the Strait of Georgia.
Frazao’s project incorporated “context awareness" into the detector by using algorithms that take the sequence of sounds into consideration, like human experts do. He also trained the algorithm on a variety of environmental cues like rain and thunder, vessel sounds, and marine mammal calls, to increase its awareness of the typical sounds in the area, and reduce misidentifications.
The top panel is a spectrogram showing a 1-minute audio recording containing orca vocalizations that was fed to the detector. The bottom panel is the detector’s output with the orange lines indicating segments of the audio recording that was in identifying the orca sounds. The green lines group those segments into a unit useful for analysis. These units could depict a single orca call.
In a sea of sounds, accurate real-time orca detectors contribute to the research into the species’ unique culture and communications, and are crucial for informed decision-making.
Frazao’s work will contribute to Humans and Algorithms Listening to Orcas’ (HALLO) development of a tool to predict the movement of the endangered southern resident orcas in the Salish Sea. More specifically, the orca detection algorithm will be integrated into HALLO’s cloud infrastructure that will be monitoring hydrophones throughout the Salish Sea in near real-time, including ONC’s Strait of Georgia hydrophones.
The data will be available through HALLO to governmental agencies like Transport Canada and Fisheries and Oceans Canada to alert vessels whenever there are orcas in the area, and to help navigate traffic around them.
Orca detection data will also be available through ONC’s data management portal, Oceans 3.0.
Southern resident orca, L105 Fluke. Credit: Jasper Kanes.
Moving forward, the orca sound detector will continue to be trained on a wide variety of environmental sounds and deployed at other ONC VENUS subsea observatory sites.
Fabio Frazao, first recipient of ONC's Roy Hyndman Ocean Observing Award in 2023.
Frazao presented his research at the Acoustical Society of America conference in Ottawa, earlier in May 2024. The findings will also be published as a peer-reviewed journal article later in the year, and included in his doctoral dissertation.
The Roy Hyndman Ocean Observing Award aims to help graduate students advance their research using ONC ocean observing data and technology, as well as work with leading ocean technology experts. The recipients of the 2024 Roy Hyndman Ocean Observing Award will be announced soon.
Learn more about the Roy Hyndman Ocean Observing Award.
Top image: Transient (Biggs) Orcas. Credit: Valerie Shore, Eagle Wing Tours.
