A dataset of Orca calls vs. other noise,
prepared by Marion Poupard and Paul Best.
Note that this is a first experiment only;
almost all files are from the same day, so
a CNN-based classifier reaches almost perfect
precision and recall (above 99%).