Analysis of underwater mammal vocalisations using time-frequency-phase tracker
Résumé
One of the most challenging applications of time-frequency representations deals with the analysis of the signal issued from natural environment. Recently, the interest for passive underwater context increased, basically due to the rich information carried out by the natural signals. Taken into account the non-linear multi-component time-frequency behaviour of such signals, their analysis is a challenging problem. In this context, the analysis of underwater mammal's whistles is aimed to extract, accurately and adaptively, their main time-frequency components. In this paper, we define a time-frequency-phase tracker which is composed of three steps. The first one consists of modelling the short-time segments of the vocalization by a set of third order polynomial phase modulations. The second step consists in the fusion of local polynomial phase modulations according to a local phase continuity criterion. Finally, in the third step, the detected time-frequency track is used to design the time-frequency filter, in charge of extracting the samples corresponding to the detected track. This procedure is then iterated until all component of interest are extracted. Tests provided for realistic scenarios and real data taken in Bay of Biscay at September 2009 containing whistles of common dolphin Delphinus delphis illustrate the potential and the benefits of the proposed approach.