Discrimination of baleen whales frequency-modulated downsweep calls with overlapping frequencies

Abstract

Spectrograms generated with the pseudo Wigner-Ville distribution (PWVD) provide much higher simultaneous time-frequency (TF) resolution compared with the traditional method using the short time Fourier transform (STFT). The WV-type spectrogram allows bioacousticians to study the fine TF structures of the sound, such as the instantaneous frequency, instantaneous bandwidth, contour slope, etc. These features set the foundation of identifying sounds that are usually considered difficult to discriminate using the traditional method. However, the PWVD requires much higher computational effort than the STFT method. In this research, the advantage of the WV spectrogram analysis was demonstrated by a case study on frequency-modulated, downsweep sounds from fin whales, sei whales, and blue whales D-calls. These calls overlapped in frequency range and have similar time duration. Automatic detection of fin, sei or blue whales FM downsweeps using the traditional spectrogram methodology tend to be ineffective because of the large temporal ambiguities needed to achieve the necessary frequency resolution. However, their WV spectrograms showed distinguishable characteristics, for example, the TF contour of fin and sei whales exhibited concave and convex shapes respectively. A support vector machine (SVM) classifier was trained and tested based on the parameters extracted from the WV spectrograms.