Bottlenose dolphins (Tursiops truncatus) produce several types of acoustical signals. One type is narrowband, frequency and amplitude modulated whistles. One of the dominant explanations for the function of these whistles has been the signature whistle hypothesis (SWH). The SWH posits that most of the whistles that dolphins produce are individually distinctive. Other dolphins can recognize the whistlers from their individually distinctive signature whistles. The SWH also hypothesizes that the contour of the fundamental frequency is the individually distinctive characteristic for signature whistles. Since the SWH was proposed, no direct experiment has been performed to test the SWH and to find which acoustical features in the whistles contain the individual identification information. The main obstacle has been the lack of tools for creating modified whistles which vary only one acoustical feature in a quantitatively controlled manner for use in the play-back experiments. Buck et al. proposed a method to synthesize and modify dolphin whistles using an assumption that the whistles consist of a superposition of harmonically-related sinusoids. We develop an alternative algorithm using the autoregressive model, which is referred to as autoregressive synthesis algorithm. The spectra of dolphin whistles show features that are appropriate to the autoregressive model, specifically strong harmonic peaks, but no deep valleys. The synthesized and modified dolphin whistles from the autoregressive algorithm sound natural, but still need more tests in the play-back experiments. 

This material is based upon work supported by the National Science Foundation under Grant No. 9733391.