We are currently working on several avenues of research relating to individually distinctive signature whistles of the Sarasota dolphins. First, we are beginning a large scale effort to create a digital library of all recordings ever made of the Sarasota dolphins. This library is completely unique, in that it contains high quality recordings of known individuals, recorded during brief capture-release events. Many individuals have been recorded on numerous occasions, thus enabling researchers to examine issues such as long-term whistle stability. A digital library will make these data much more widely accessible to other researchers.

UNCW undergraduate student Charles White continues to work on an automated classifier for bottlenose dolphin signature whistles. Currently, his study is focusing on whistle detection, by comparing the performance of a k-means clustering approach to that of a feed forward neural network. While both approaches successfully detected dolphin signature whistles among noise, the neural network was more robust in handling irregularities such as whistle tail-offs and recording dropouts. Results so far provide a reliable method for autonomously detecting bottlenose dolphin signature whistles and lay the groundwork for automated extraction and classification of whistles.

We also use the Sarasota whistle catalogue to develop technologies for signature whistle identification in the field. Currently, the only way to identify a signature whistle is to record an isolated individual. However, animals usually travel in groups. It would be very useful to be able to recognize a signature whistle during a boat follow without any previous information on the animals in the group. For this, we developed a computer method that can categorize dolphin whistles automatically. It consists of an adaptive resonance theory neural network that is unsupervised in its learning phase. This program has been demonstrated to be successful with small sample sizes. We are now testing its performance with larger data sets from the Sarasota dolphins to adapt it for field work use.

We also are continuing our playback studies, which are designed to determine the cues that dolphins use in recognizing signature whistles of other individuals. As described in last year’s newsletter, experiments conducted in 2003-2004 showed that dolphins are capable of recognizing synthetic signature whistle contours, suggesting that contour is the most important feature of the whistle for individual recognition (Janik et al. in prep). However, these experiments did not rule out the possibility that dolphins use both contour and voice cues to recognize individuals. Thus, our current experiments are looking at whether dolphins are capable of discriminating among natural non-signature whistles. If they are capable of discriminating among these whistles, which are highly variable in contour, then they must be using voice cues for this recognition. Preliminary analysis of ten playbacks showed no difference in responses to non-signature whistles of kin vs. non-kin. These preliminary results suggest that voice cues are not used by dolphins to identify whistles of other individuals. This work is currently being funded by a Protect Wild Dolphins Grant to L. Sayigh and R. Wells from the Harbor Branch Oceanographic Institute, and a Royal Society University Research Fellowship from the UK to V. M. Janik, with additional support from Dolphin Quest and Disney’s Animal Programs.