Cetacean vocalization tracking using an adaptive Poisson rate estimator

Abstract

This paper presents algorithms for tracking the formant structure of cetacean vocalizations using auditory modeling and adaptive filtering to estimate the Poisson intensity of nerve firing rates. These rates are in turn related to the instantaneous frequency of the formant vocalisation. The auditory model considered in this paper produces a process described as a spike train representing the auditory nerve firing, the Poisson intensity of which is determined by an underlying modulation process. Previous work [Kenny et al., Adaptive Filter for Speech Enhancement Using Poisson Rates from an Auditory Model, paper presented at the WESPAC VIII, Melbourne, Australia (2003); C. McPherson, Development of a Toolkit for Analysing and Localising Underwater Acoustic Signals (James Cook University, Townsville, 2005)] has shown that this modulation process can be evaluated from the conditional expectation of the Poisson rate given interspike arrival time observations. The paper also addresses the application of adaptive filters to update the estimator parameters for different acoustic environments, which is unknown. These algorithms have been applied to real field recordings of interacting false killer whales (Pseudorca crassidens) and results from experiment of shown individualisation and extraction of individual narrow-band vocalizations.