Mapping the phonetic structure of humpback whale song units: extraction, classification, and Shannon-Zipf confirmation of sixty sub-units

Abstract

The striking similarities of time-frequency spectrograms of voiced human speech and humpback whale vocalizations suggested a common targeted frequency-modulated phonetic basis. To map the sub-unit structure of humpback whale song units, a time-frequency contour segmentation, extraction, and classification procedure was developed and tested on streaming voiced human speech. When the procedure was applied to humpback vocalizations and the tone-pairs of the two most energetic “vocal fold” harmonic frequencies were plotted in x-y coordinates, the plot exhibited properties of an optimally structured Shannon “modem symbol constellation” diagram of 14 distinct sub-regions and 60 acoustically distinct sub-unit symbols. The humpback symbol constellation is structurally comparable to the tone-pair symbol constellations of English and Asian language vowels. The information entropy and plot of the humpback sub-unit symbol set’s cumulative probability vs. ranked frequency distribution function are nearly identical to the entropy and Zipf power-law profile of the English language phoneme set. The precise specification of the sub-unit structure of more than one hundred song units analyzed to date has resulted in a significant expansion in the number of unique units identified in previous studies, suggesting that the “lexicon” of humpback song units is potentially much larger than cited in the literature.