Baleen Protocols

Low-Frequency Detection and Classification System (LFDCS) Baleen Whale Detector and Classifier

Low-frequency tonal detector for baleen whales; written by Mark Baumgartner, see Baumgartner and Mussoline 2011


FIG. 1. A pitch-tracking example. (a) Spectrogram [S; Eq. (1)] created from short time Fourier transforms of the audio data (sampling frequency¼2048 Hz, frame¼640 samples, overlap¼80%, Hann window) and smoothed with the smoothing operator [M; Eq. (2)]. Four calls are present: two sei whale downsweeps (40–100 Hz, 1.4 s duration) and two right whale moans (120–170 Hz, 2.7 s duration). (b) Filtered spectrogram [A; Eq. (3)] created by subtracting a running mean from each discrete frequency band. Note removal of 220–225 Hz tonal noise. (c) Pitch tracks of calls with average amplitudes in excess of 12 dB relative to background. (d) Same pitch tracks in (c) with amplitude displayed in color. Baumgartner & Mussoline, 2011

Humpback Whale Convolutional Neural Network (CNN) Detector

Humpback whale convolutional neural network detector; originally developed by Ann Allen/Google and modified for the North Atlantic humpback whale population by Vincent Kather

  • Makara detector code: PYTHON_HUMPBACK_CNN
  • Github

Ketos Minke Whale Detector

Atlantic minke whale pulse train binary ResNet18 deep neural network (DNN) detector (currently in use); developed by Xavier Mouy