Configuring a Whistle Detector
Whistle Detector
Configuring the Whistle Detector
Parameters for each of the three main stages of whistle detection are accessed via the Detection>Whistle Settings menu.
FFT Data Source
A drop down box will show a list of available data sources for the whistle detector. These are either the output from an FFTEngine, or some other process that produces FFT / Spectrogram data. If no data sources are listed, you need to create an appropriate FFTEngine in the PAMGuard data model.
Channel List and Grouping
Select which channels you wish to detect whistles on and which pairs or channels should be used to calculate bearings to detected whistles.
To do this, select the channels you wish to use using the check boxes, then assign group numbers. Only the first channel in each group will be used for detection, if a whistle is detected, then data from the second channel in the group will be used to calculate a bearing to that whistle.
In the example above, whistles will be detected on channels 0 and 2. Should a whistle be detected, then channel 1 or 3 will be used to calculate a bearing.
Peak Detection
Two peak detectors are currently available. You are recommended to use the ‘IFAW Whistle Peak Detector’ which is the same as the peak detector on the old IFAW whistle software.
The first stage of the whistle detection process is to compare the instantaneous values in the spectrogram S to a measured value of background noise. The background noise N at time t and frequency f is measured using
where the constant α is given by α=Δt/τ,
Δt being the time between successive time partitions in the spectrogram and τ is the smoothing constant set in the Peak Detection dialog. Two values of the smoothing constant are used, one when the data are below threshold which allows relatively rapid response of the system to varying noise levels and one when the data are already above threshold to avoid the presence of a whistle affecting the chances of detecting further whistles by raising the detection threshold.
After background noise measurement, the total number of frequency bins in each time partition that exceed the detection threshold is counted. If this exceeds a set percentage of the total bins (Maximum bins over threshold) it is assumed that there is a broad band click and no whistle peaks are extracted from that time partitions. Otherwise, groups of bins in frequency that exceed threshold and lie between the specified minimum and maximum frequency width are used to create spectral peaks which are passed to the whistle linker.
Linking
Three parameters control whether or not peaks from different time partitions can be joined into putative whistles.
Several parameters can be specified.
Max Sweep: the maximum allowed frequency sweep rate.
Max Sweep Differential : the rate at which the frequency sweep can change (how curved the whistle contour can be)
Max Amplitude Change: the rate at which the peak amplitude can change along the length of the whistle.
For a peak to be added to the whistle, all three criteria must be satisfied (except on the very first linkage for each whistle, in which case the differential term is not used). If more than one peak is a candidate for linking to a whistle, then an overall ‘goodness of link’ value is calculated using the weight terms for each parameter.
Max gap is the maximum number of time partitions that can be skipped at any point within a whistle with peaks still being linked.
Whistle Selection
Whistles are only created if the the linked peaks (putative whistles) pass two final tests:
Minimum Length: Whistles must exceed some minimum length
Minimum Occupancy: If the Max gap parameter in the Linking stage is > 0, then whistles are only created if the number of peaks making up the whistle occupy a minimum percentage of the total whistle length.
Event Detection
A Whistle Event occurs when the rate at which whistles are produces exceeds some minimum value. Since there is generally a low rate of false detections of individual whistles and since most whistling species produce many whistles, whistle events are often a more useful indicator of the presence of animals than individual whistles.
Integration Time - time to count whistles over.
Minimum Whistle Count - minimum number of whistles required to form a whistle event
Maximum gap - time interval after the last whistle in an event before the event is closed and counting starts again from 0.