Project

Speaker Identification of Marine Mammals

Speaker identification techniques (such as a Gaussian Mixture Model (GMM) and a Hidden Markov Model, using cepstral coefficients as features) have been applied to determine the question of whether individual marine mammals can be identified by their vocalizations alone. With a dataset of four killer whales uttering sounds previously classified as belonging to call type n2, and over 10 sounds from each individual, we have found a very high success rate of 80 to 100% correct for the six pairwise comparisons and around 78% correct for identification among all four individuals. The ability to identify marine mammals from their vocalizations alone, in addition to the theoretical interest for production mechanisms, is extremely valuable in the ability to track these mammals from remote locations where visual information is not present.

Speaker identification techniques (such as a Gaussian Mixture Model (GMM) and a Hidden Markov Model, using cepstral coefficients as features) have been applied to determine the question of whether individual marine mammals can be identified by their vocalizations alone. With a dataset of four killer whales uttering sounds previously classified as belonging to call type n2, and over 10 sounds from each individual, we have found a very high success rate of 80 to 100% correct for the six pairwise comparisons and around 78% correct for identification among all four individuals. The ability to identify marine mammals from their vocalizations alone, in addition to the theoretical interest for production mechanisms, is extremely valuable in the ability to track these mammals from remote locations where visual information is not present.