Avisoft logo

SASLab Manual

Spectrogram window : Tools > Automatic parameter measurements > Automatic parameter measurements setup > Element separation

The element separation (segmentation) can be accomplished by magnitude thresholds, spectral property thresholds or manually by selecting the desired elements by section labels:

automatic (single threshold) Applies a single magnitude threshold to the spectrogram to both identify the individual elements and locate their temporal boundaries.

This simple option can be used when the individual elements exhibit uniform amplitudes.

automatic (two thresholds) Applies a primary Threshold to identify the individual elements and an additional start/end threshold to locate their temporal boundaries.

This slightly more complex option can be used to safely detect elements exhibiting varying amplitudes.

start/end threshold This additional threshold is only available on the two-thresholds element separation mode. The start/end threshold is referenced to the maximum magnitude of each detected element. The start and end of the element are those points in time where the magnitude amplitude falls below the start/end threshold relative to the element maximum (for at least the Hold time). In this way, the start and end points of elements with varying amplitudes will be recognized more precisely. The threshold should normally be set to about –20 ... -10 dB, depending on the structure of the elements and the background noise level. In recordings with a high level of background noise, this secondary threshold should not be set too low. The + and – buttons can be used to interactively find the best settings.

automatic (three thresholds) This option is very similar to the above two thresholds option, except that two separate thresholds are being used for detecting the start and the end of each element.

This option might be useful for detecting bat echolocation sweeps that start softly at high frequencies and end at higher amplitude with a reverberation tail. Due to reverberation tails, the above two thresholds option may fail to detect both the start and end of the calls precisely. The three thresholds mode can be adjusted in such a way (with a low threshold for the start and a higher threshold for the end) that such calls can still be detected reliably.

Threshold If this threshold is exceeded, the start of a new element is registered. When the signal goes below that threshold for at least the Hold time, the end of that element is registered. In order to find the best threshold settings for the two or three thresholds mode, the single threshold mode should be selected first. This would allow finding an optimal setting for this primary threshold. The threshold is appropriate once all elements are recognized properly regardless whether the start and end points are correct. The secondary start/end threshold can then be adjusted after switching into the two (or three) thresholds mode. The show threshold option allows adjusting the threshold visually by dragging its graphic representation above the spectrogram.

The option monotonic decreasing will accept only monotonously decreasing frequency courses in an element, which can help rejecting echoes in bat calls.

relative to maximum If this option is activated, the primary Threshold is referenced to the maximum overall magnitude value of the entire spectrogram. Otherwise, the Threshold dB value is interpreted as an absolute dB(FS) value referenced to the full scale (0 dBFS corresponds to the full scale = 1V peak amplitude) or as an absolute dB SPL value if the sound file has been calibrated (main window command Edit > Tools > Calibration…). Use this option to analyze recordings with varying signal amplitudes due to varying distances between the vocalizing animals and the microphone.

show threshold This option activates a graphic representation of the element separation threshold that may help adjusting the threshold.

total energy If this option is activated, the threshold comparison is based on the root mean square (total energy) of the spectra rather than the default peak amplitudes (magnitudes) of the instantaneous spectra. The total energy option should be activated for analyzing multi-harmonic or noisy sounds because the peak amplitude in such signals is often too low and too variable for a save element detection.

interactively (section labels) If activated, manually entered or manually edited section labels are being used to identify the individual elements.

The option number labels automatically will assign consective number to the label texts.

The layer list box allows to filter the section labels used for the Automatic Parameter Measurements by their layer attribute. The default option all layers will include all section labels regardless of the layer.

This type of element separation can be useful in noisy recordings that prevent a reliable automatic detection. Nevertheless in can often be useful to first use an automatic separation and subsequently applying the command Copy detected elements into section labels or clicking at the edit> button. These automatically detected labels could then be modified manually to correct inaccurate results.

automatic (whistle tracking) This element separation method is useful for detecting soft whistle-like vocalizations in noisy recordings (certainly for analyzing USV’s by laboratory mice and rats).

The implemented algorithm searches for steady signals having a relatively stable (peak) frequency course without rapid frequency modulations.

Principle of the whistle tracking mechanism

The amount of frequency modulation that should be tolerated can be selected from the edit field max change. This is the tolerated maximum change of the peak frequency between two consecutive time bins on the spectrogram, expressed as the number of pixels on the spectrogram (frequency bins). It therefore depends on the currently selected spectrogram parameters (FFT length and overlap) and should be set to a value between 1 and 4 pixels (depending on the time and frequency resolution of the spectrogram). For best results, this parameter should be set to the smallest value that still properly recognizes all the elements to be detected. The whistles button in the Presets section on the Automatic Parameter measurements setup dialog box provides appropriate settings to start with.

The slope option allows specifying the minimum and maximum frequency change (slope) independently, making it possible to detect also frequency-modulated calls.

The min duration setting rejects any whistles that are shorter than the specified duration. In order to prevent false positive detections (caused by noise), this value should be as large as possible (but shorter than the duration of the vocalizations that should be detected). It is recommended to set the Hold time parameter temporarily to zero while optimizing the max change and minimum duration settings. The option show trace will indicate the peak frequency at each time bin that is being used by the whistle tracking algorithm.

The monotonic option will only accept monotonically rising or falling peak frequency shapes, which can improve the efficiency of the algorithm in distinguishing whistle-like calls from noise.

Unlike the other element separation options, the whistle tracking algorithm itself does not use level thresholds. Nevertheless, it is still possible to filter out unwanted soft calls or noise by activating the option Reject if peak ampl. < xxx dB located at the bottom of the dialog box.
In case there is low-frequency noise or a DC offset present in a recording, the whistle tracking mechanism may fail to properly detect soft vocalizations. This problem can be fixed by limiting the frequency range of the spectral analysis from the spectrogram window command Display > Cut-Off Frequencies… (enter here an appropriate high-pass frequency in order to exclude the disturbing components).

automatic (spectral entropy) This element separation method employs the Wiener Entropy that can be used to detect tonal vocalizations in noisy recordings.

The Wiener Entropy (spectral flatness) is calculated by dividing the geometric mean of the instantaneous spectrum by the arithmetic mean of the spectrum and ranges theoretically from 0 (pure tone signal) to 1 (white noise). An element is detected each time when the entropy falls below the defined max entropy threshold. In order to exclude unwanted short noise signals, the min duration threshold on the Post filter on elements section can be used (set the max entropy threshold to 1). Use the show trace option to visualize the trace of the entropy.

automatic (spectral bandwidth) This element separation method employs a spectral bandwidth parameter that can be used to detect tonal vocalizations in noisy recordings.

The spectral bandwidth parameter is calculated by counting the n most intense frequency bins of the spectrum whose integral represent a user-defined fraction of the total energy of the instantaneous spectrum. An element is detected each time when the calculated spectral bandwidth falls below the defined maximum number of frequency bins threshold. Note that the frequency bins threshold will be affected by the analysis bandwidth of the spectrogram that is determined by the FFT size and the windows type. In order to exclude unwanted short noise signals, the min duration threshold on the Post filter on elements section can be used (set the max entropy threshold to 1). Use the show trace option to visualize the trace of the spectral bandwidth.

The edit> button creates section labels from the automatically detected elements and switches to the interactive element separation mode. That way it is possible to edit the automatically detected elements, making it possible to label undetected elements. New section labels can be inserted by pressing the Shift key and left-clicking at the desired start point in the upper third of the spectrogram and then dragging the mouse cursor to the end point. The section labels can be resized and shifted using the mouse cursor or deleted by right-clicking at them while the Shift key is being pressed.

Hold time In order to prevent cutting up integral elements, a hold time can be specified. If the amplitude (or the spectral criterion) of an element falls below the threshold for less than this hold time, it will still be treated as a single element only (otherwise two separate elements would be detected). This parameter must be adjusted carefully in order to get the desired results.

In order to exclude low frequency noise signals, an appropriate high-pass cut-off frequency can be specified from the spectrogram window drop-down command Display > Cut-Off Frequencies....

Avisoft Bioacoustics last modified on 05 October 2022