A time-frequency perceptual feature for classification of marine mammal sounds

Marine mammal sounds exhibit distinctive features associated with species, individual identity, and certain behaviors, and automatic recognition of marine mammal become an interesting and important approach for acoustic investigations of free-ranging animals in their natural habitats. The acoustic object classification using auditory perceptually features are a promising approach developed recently, and among the auditory features the loudness is a basic and key one. Thus, calculation of the loudness of marine mammals sound is investigated in this paper. Because marine animal vocalization is dependent strongly on time, traditional auditory model for steady sound cannot be used and thus the focus is on the auditory perception model for transient sounds. Main ideas are firstly to use STFT approach to decompose temporal signal according to maximum power changing with time in frequency domain, then calculate specific loudness in time-frequency domain according to recently revised Moorepsilas model, which is a temporal integration of specific loudness modified by masking characteristics. Finally specific loudness in time-frequency domain is obtained for numbers of vocalizations from two species of baleen whales.