Previous Page 
|
Previous Page |
Next Page |
Signal detection is used to find the candidate vocalizations within a recording. These candidate vocalizations can then be compared against vocalizations of interest using recognizers. The goal of signal detection is to locate candidate vocalizations in potentially noisy recordings and determine where the vocalization begins and ends.
A typical vocalization consists of a series of "syllables" tightly grouped together into a "song" spanning only a few seconds in total duration or less. Syllables are usually only a fraction of a second in duration, and the inter-syllable gaps between syllables are also only a fraction of a second in duration. Sometimes the inter-syllable gap is so small that multiple syllables may appear to merge together.
To see a visual representation of Song Scope's signal detection at work, you can display the waveform plot using Logarithmic Scale with Signal Detection. This view shows the total power levels within the selected frequency range as set by the Frequency Minimum and Frequency Range spectrogram controls. Different colors are used to indicate the results of Song Scope's signal detection algorithm:
No Signal
Portions of the recording that are not considered to be part of a candidate vocalization are shown in colors used to represent relatively weak signals on the spectrogram (a dark blue by default).
Syllable
Portions of the recording considered to be part of a syllable within a candidate vocalization are shown in colors used to represent the strongest signals on the spectrogram (red by default)
Inter-syllable Gap
Portions of the recording considered to be part of an obvious inter-syllable gap are shown in colors used to represent medium signals on the spectrogram (a green by default).
Soft Gap
Portions of the recording considered to represent a likely inter-syllable gap where two syllables merge together are represented with a color used to represent signals between medium and strong (yellow by default).
Note that recordings with weak signals or low noise may not register any signal above 0dB for color-coding. You can increase the gain by adjusting the Gain Control to bring these signals into view.
Also note that you can adjust the display colors using the Display Controls
The following controls can be used to tune the signal detection algorithm for optimum results when looking for a particular candidate vocalization:
The signal detection algorithm looks at total power in the visible frequency band as determined by the Frequency Minimum and Frequency Range controls in the Spectrogram Control Panel. In a typical audio recording, there is significant noise in lower frequencies that make it difficult to accurately detect signals unless they are filtered out. You should therefore adjust the minimum frequency as high as possible without clipping the lowest frequency component of the vocalization of interest. By limiting the upper frequencies, you can avoid triggering the signal detection when vocalizations of a higher frequency animal are heard.
The Max Syllable control on the detector control panel is used to specify the largest syllable likely to be encountered in the vocalization.
The Max Syllable Gap control on the detector control panel is used to specify the largest inter-syllable gap likely to be encountered in the vocalization. If a "quiet" interval exceeds this value, then the Song Scope detector will mark the end of the vocalization. If this value is too small, the detector may not group all the syllables of a song together. If this value is too large, it is possible that a second vocalization from a different and unrelated individual may be incorrectly joined with the vocalization of interest.
The Max Song control on the detector control panel is used to specify the largest vocalization likely to be encountered in the vocalization.
The Dynamic Range control on the detector control panel is used to limit the dynamic range of the relative power levels in the vocalization as part of Feature Reduction. In addition, dynamic range plays a role in signal detection in cutting off weaker vocalizations in favor of selecting stronger ones for candidates. If this value is too low, there will not be enough information to detect important elements of the vocalization for accurate recognition. If this value is too high, the signal detector and recognizers will be more susceptible to background noise. The optimum value for the dynamic range control is the expected signal to noise ratio of the field recordings. That is, the difference in decibels between the typical background noise and candidate vocalizations within the specified frequency band.
The Algorithm control on the detector control panel is used to select between the new and improved version 2.0 classification algorithms and the older version 1.0 algorithms (for backward compatibility with older recognizers or in case the older algorithms perform better for some applications).
|
Previous Page |
Next Page |