Manual identification
Manual identification was used to identify candidate focal species and
collect vocalizations to be used in species classifiers. A random
selection of 1,000 audio files from the complete audio dataset were
manually assessed and a record was made of all species that could be
positively identified within each file in order to determine which
species were heard most often and would therefore make the best
candidate focal species. Files with positive identifications were then
used as part of the training datasets for the single species
classifiers. These manual identifications were used again later in the
workflow to measure the recall rate of the completed species
classifiers.