4
Appendices
4.1
SoundTrap HF Click Detector
The SoundTrap ‘HF Click’ Detector is a general purpose high frequency click detector followed by a
snippet extractor. The detector is designed to detect most odontocetes and is intended to be used
as a first pass, guiding you to times in your recordings with lots of transients. You would typically use
the detector with a low detection threshold so that it makes a lot of false detects but also detects
most genuine clicks (i.e., high sensitivity, low specificity). You then evaluate the reported detections
in Pamguard to reduce obvious false detects, to classify clicks into species groups, and to identify
bouts, as you would with a continuous wideband recording. The benefits of doing on-board click
detection are that you get much longer recording times from your SoundTrap and you can identify
times of interest more quickly when you get the data back. The downside of course is that you do
not have the full wideband recording on the SoundTrap. This is mitigated in SoundTrap’s detector by
recording a short snippet of wideband sound around each detection. A few 100's of µs is often all
that is needed to classify odontocete clicks into general classes. The detector can be configured to
also make a continuous recording at lower bandwidth that allows you to quantify the general
ambient noise conditions, an important factor affecting detector performance, as well as the
presence of low frequency whistles that can help with species identification.
4.1.1
Click detector
The click detector comprises an approximate pre-whitening filter followed by a power-in-band
Constant False Alarm Rate (CFAR) detector. The whitening filter (a 35kHz high-pass filter) corrects
the typical low-frequency emphasis of underwater ambient noise creating a more spectrally-
balanced signal for the detector. The power-in-band detector consists of a highpass Finite Impulse
Response (FIR) filter (approximate 3 dB bandwidth 115 - 160 kHz) followed by a moving average
power estimator. The duration of the moving average window is a user parameter (see below). In
parallel, the power of the full bandwidth whitened signal is measured using a much longer
exponential window to give an estimate of the ambient noise power. This is used to adjust the
absolute detection threshold: a detection will be made if the power-in-band is greater than the
ambient power by more than the Relative Detection Threshold (RDT) which is a user parameter.
When a detection is made, the time of the detection and the ambient noise power at the time are
passed to the snippet extractor. A user-selectable blanking time must then elapse before another
