A New Approach for North Atlantic Right Whale Upcall Detection

In this paper, a new up-call detection algorithm for North Atlantic Right Whales is proposed. Mel Frequency Cepstral Coefficients and Discrete Wavelet Transform (DWTs) are used to extract features of North Atlantic Right Whale up-calls. Several types of wavelets are tested in terms of detection accuracy. Once the up-call features are extracted, classifiers such as Support Vector Machines and KNNs are applied to classify call types. Detection results shows that the upcall detection rate by using MFCCs is 73.82%. However, the upcall detection rate is improved to 92.27% when MFCCs are integrated with DWTs for feature extraction. Furthermore, it is shown through experimental studies that the proposed detection scheme is superior to those obtained using spectrogram-based techniques in terms of both detection accuracy and speed.