Orcas, also known as Killer Whales or Blackfish, pose a challenge for researchers due to their underwater travels in packs and the difficulty of identifying individual whales. OrcaLab, a research facility on Hanson Island off the northeast coast of Vancouver Island, has tackled this issue by installing underwater hydrophones to track Orcas based on their vocalizations. However, identifying specific individuals within groups remains a challenge as there are no distinct patterns in vocal sounds.
To try and address this problem in the 1990s, Praxis proposed using a Convolutional Neural Network (CNN) and machine learning to categorize whale voice recordings into groups belonging to each whale. The analog recordings were digitized, and the CNN system was used to sort the data, resulting in nearly thirty different audio collections representing different Orca individuals.
Using this data, vocal “models” were created for each individual, enabling the identification of whales based on new vocalization recordings, even without visual identification. Over time, the sorted whale vocalizations, paired with identification numbers, allowed for the visual identification of certain individual whales in correlation with their vocal identifications. This research received partial funding from the National Research Council of Canada.