Bottlenose dolphins communicate acoustically using transient signals known as whistles. Recordings of this form of communication usually contain unrelated signals such as noise and echolocation clicks in addition to the desired whistles. Current methods of analyzing these recording include manually extracting whistles using spectrograms. An automatic whistle detector will allow processing large volumes of recorded data efficiently. Signals of interest are detected by comparing the average energy of the bandpass filtered signal against a threshold. Additional hysteresis is applied using timing information from the raw energy detections. The energy thresholds are chosen manually for each recording processed and are set such that the probability of false alarm is larger than the probability of a missed detection. The automated detections are then screened by a human user examining spectrograms calculated for the potential whistles. The true whistles confirmed by the user are then saved in individual files along with the animal's name, recording date, track name, and the start and stop times for each whistle.

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