“Measurements”. This tab contains settings for how you measure acoustic units on the spectrogram. There are two types of acoustic unit in Luscinia: Elements and Syllables.
To measure elements, the user clicks inside the spectrograph (while in Element mode), and moves the cursor to highlight a contiguous region of the spectrograph. Progress is recorded as a semi-opaque red blob that is superimposed on the spectrograph. The aim is to highlight the element that is to be measured by ensuring that the red blob covers the element. It is not essential, if the recording is good, that the red blob covers ONLY the element – Luscinia possesses a simple algorithm to search for a signal inside the red blob. Moreover, if the elements are separated, you can highlight multiple elements at once, and Luscinia will then segment them. For more on this algorithm, see here. However, if the recording is noisy, care should be taken not to highlight any artifacts that Luscinia could mistakenly include in the measurements. Once the element is highlighted, the user should (left-)click the mouse again, and the red region should turn into one or more green elements.
To measure syllables, the user clicks once inside the spectrogram (while in syllable mode), moves the mouse until all the elements that should be included in the syllable are included, then clicks again.
The Mode button changes the mode from measuring elements to measuring syllables and back again.
The Undo button undoes what ever element change had been made most recently.
The Re-do button redoes whatever might have been just undone by the Undo button.
Select all is equivalent to highlighting the entire screen. This option is only useful for very good quality recordings.
Re-measure recalculates the measurements made for the last measured element. This option is useful if you highlight the element correctly, but Luscinia does not correctly measure some aspect of the element (e.g., fundamental frequency, or how the highlighted area is segmented into different elements). In such a situation, you can vary the relevant parameter (e.g. increase dynamic range), adjust the spectrogram, then click on Re-measure.
Calculate syllables automatically is a time-saving function that allows the user to ask Luscinia to automatically estimate the hierarchical syllable structure of a song. In many cases, syllables are apparent due to repetitions, or clear patterns of temporal rhythm (gaps between elements within syllables being much shorter than those between syllables). You can use this function to try to automatically estimate syllable structure. Luscinia uses its Dynamic Time Warping algorithm to compare elements within the song with each other. More information on the DTW parameters can be found here.
Delete element allows you to delete elements from the list of elements associated with the sound.
Delete syllable allows you to delete syllables from the list of syllables associated with the sound.
Merge elements instructs Luscinia to merge two neighbouring elements. Occasionally Luscinia may split one element into several due to a quiet spot in the recording; this option allows a way to rectify such an error. Luscinia simply carries out a linear transformation to match up the edges of the two elements, and searches for the highest intensities within that space. This option is useful in the case where the boundaries between elements are somewhat ambiguous.
FF jump suppression. In the process of estimating Fundamental Frequency, a post-processing phase is to go through the element and remove frequency jumps - points where the fundamental frequency estimate rapidly jumps from one frequency to another. Such jumps are often caused by the algorithm jumping to from the fundamental to the first harmonic for example. Setting this parameter controls the degree to which rapid jumps in frequency are allowed. Jumps are suppressed more heavily if a low number (e.g. 0.1) is entered.
FF bias is a second parameter associated with measuring fundamental frequency. Fundamental frequency is the most difficult of the element measurements that Luscinia makes – it can easily be thrown off by noise. FF bias simply changes the way that different hypotheses about the FF are weighted: a high value will tend to decrease the measured Fundamental Frequency.
Min. gap tells Luscinia what the minimum gap between elements should be when an area of the spectrograph is highlighted. Once Luscinia has calculated new elements based on what a user has selected, it consults this parameter. If two of the new elements are less than this distance apart, Luscinia will merge them.
Brush size sets the size of the selection tool used to highlight elements. Setting it to a large size allows rapid measurement of a sound (for clean recordings). Setting it to a small size allows precise, but slower measurement.
Upper hysteresis loop and Lower hysteresis loop. The algorithm that Luscinia uses to detect elements inside the red blob is adapted from a commonly used edge-detection algorithm in computer graphics, using a hysteresis loop. Luscinia searches through the area of the spectrograph that has been highlighted (the red blob). If the program comes upon a point in the spectrograph whose intensity is above the upper threshold, then that point is included as a signal. Then, Luscinia branches out in all directions around this point, and includes any point whose intensity is above the lower threshold. This process continues until the intensity of all points around this growing area fall below the lower threshold, or until the edge of the highlighted region of the spectrograph has been reached. The algorithm then continues to search through the rest of the highlighted area for points with intensity above the upper threshold to start the process again. Note: if this process results in distinct regions of signal that have a gap in time between them, then Luscinia will create more than one new element.
Min. length sets the minimum length for elements. Detected elements that fall beneath this threshold are discarded. If this is set to a small value, Luscinia is more likely to detect noise artifacts and record them as elements.