Recogniser construction
To construct the recognisers, we used the technique ‘binary template
matching’ – a technique that first converts a spectrogram into a binary
template and then matches ‘on’ and ‘off’ points of the template to the
file the recogniser is applied to (Katz et al., 2016b; Towsey et al.,
2012). This was done using the monitoR package in R, which provides
flexibility in its construction parametres (Katz et al., 2016b). To
build the initial recognisers, templates were constructed from a minimum
of 10 reference calls (from the pool of 100 candidate calls) that were
both clear and representative of the variation in calls and
environmental conditions. Binary templates comprise ‘on’ and ‘off’
regions (call and non-call), which are based on a user-defined amplitude
cut-off (Figure 1). Each template’s amplitude cut-off was determined
through manual inspection of templates using the makeBinTemplate
function in monitoR (Katz et al., 2016b). Amplitude cut-offs were set
arbitrarily and progressively altered and reviewed. A cut-off that
clearly showed call structure and was not masked by background noise,
was deemed appropriate, with some background noise deemed acceptable.