A method for tracing dolphin whistles

Abstract

Underwater acoustic recordings containing dolphin vocalizations are often analyzed in time-frequency domain using spectrograms. Spectrogram feature extraction techniques are widely adopted in whistle classification studies because they provide a visual representation of the whistle's frequency variation over time. However due to the low SNR of recordings, harmonics and frequency spread, most researchers use time consuming manual methods to trace whistles. The work presented in this paper attempts to automate this process. © 2006 IEEE.