Blue whale B and D call classification using a frequency domain based robust contour extractor

Passive acoustic monitoring of blue whales (Balaenoptera musculus) has been used to gain insight into their presence and behavior. Many of their calls have been shown to be detectable through spectrogram correlation due to the low variation in these stereotyped calls. In this work, we describe rule based classifiers for tonal B and D calls using the pitch/frequency contour information obtained from a contour extractor. B calls can be detected by spectrogram correlation, but the D calls are highly variable and are therefore difficult to detect using spectrogram kernel methods. Experiments on four hours of evaluation data from different field seasons show that 91.3% of B calls and 85.8% of D calls were correctly retrieved. For both types of calls, less than 2% of the retrieved calls were false positives.