To design effective alarms, you need to look for patterns in the types of bat calls
and types of non-bat noise that any particular microphone is recording.
Procedure
-
Vet your recordings to identify bat species and also mark any recordings that
contained non-bat noise.
-
Using a spreadsheet processor, look for patterns in call parameters that could
be used to distinguish bats from noise.
Imagine a common type of noise in your recordings produces shorter pulses
than low-frequency bats, but narrower bandwidth than high-frequency bats.
One strategy for distinguishing between bat calls and noise would be to
create separate alarms for low- and high-frequency bats.
The low-frequency alarm would look for sounds only up to a certain maximum
frequency with pulses longer than a certain duration.
The high-frequency alarm would look for sounds above a minimum frequency,
wider than a specific bandwidth.
What to do next
Design alarms to take advantage of the patterns you have found in your data. Then,
test how effectively they trigger for bat activity without triggering for
noise.