Abstract
The signature whistle hypothesis proposes that most of the whistles bottlenose dolphins (Tursiops truncatus) produce are individually distinctive, and that bottlenose dolphins can identify the signature whistles of the other dolphins from the fundamental time-frequency contour. This hypothesis has not been tested directly. One major obstacle has been the need to modify the whistles so that scientists can quantitatively control the acoustic parameters of the whistles. Buck, Morgenbesser, and Tyack [J. Acoust. Soc. Am. 108 (2000)] proposed a model for synthesizing whistles using a sum of harmonically related sinusoids. Some synthetic whistles produced by this method sound artificial. The objective of the current research is to develop an alternative technique for modifying and synthesizing these whistles using autoregressive (AR) modeling. An important property of the AR model is that if the synthesis filter is excited by the residual signal from a prediction filter, the original signal is perfectly reconstructed. Using the residual signal as the excitation for a synthesis filter with modified AR parameters should produce more realistic-sounding modified whistles. [Work supported by NSF Ocean Sciences.]