Extraction of auditory related features for marine mammal recognition

Marine mammal recognition has application value on various fields. In the complex underwater acoustic environments the recognition performance is influenced by many factors such as noise and channel effect. In this paper, combining Hilbert-Huang transform and Gammatone filter bank, an auditory related time feature has been proposed to enhance robustness. Hilbert-Huang is used for the time frequency analysis and Gammatone filter bank is used for the simulation of auditory perception. Vocalization data-set of four types of whales were used for the recognition experiment. The results are compared with those of MFCC in various SNR conditions and have validated the efficiency of the proposed feature.