Preparing sounds
Fast Track is intended to analyze sound files that contain only a single vowel sound, or vowel nucleus.
The method used to pick the best analysis assumes that formants exist and are continuous throughout the recording. As a result, Fast Track can also be used to analyze sound files that contain sequences of sonorants and vowels, or potentially other sounds, provided that these feature formants without discontinuities, and that the order of the regression analysis is large enough to adequately represent the movement of the formants in the interval.
If files feature portions where no formants are to be found, results may be unpredictable and bad, since they will be based on the 'best' formant tracks across sections with no formants.
One thing to keep in mind is that Praat will 'secretly' use an analysis window that is twice as long as the one you specify:
"Window length (s) the effective duration of the analysis window, in seconds. The actual length is twice this value, because Praat uses a Gaussian-like analysis window with sidelobes below -120 dB. For instance, if the Window length is 0.025 seconds, the actual Gaussian window duration is 0.050 seconds."
This means that if use a window length of 25 ms, you are actually using a 50 ms window. At the moment Fast Track only uses 25 ms anlyses (50 ms windows), but if this parameter were changed, padding would have to change accordingly.
A formant analysis can never extend from time 0 to time x (for a sound that is x milliseconds long), because of the length of the window used to carry out an analysis. If the analysis window is 50 ms long, that means that formant estimates will extend from 25 ms (windowlength/2) to x-25 ms.
For example, say a researcher is interested in tracking the formants in the 100 ms between the green lines. If you extract only that, you will only get formant values between the read lines. However, if you extract from 0 to 150 ms (the whole image), you can actually get formant measurements right up until the green boundaries (the measurements determined to be 'in' the vowel).
Fast Track offers a few ways to deal with this.
-
Use the tools in Fast Track to pad sound files during extraction from TextGrids. For example, if the vowel sound starts at 325 ms in the recording, the sound is extracted starting at 300 ms. This allows the first formant estimate to be at 325 ms, correspond to beginning of the segment.
-
Accept the limited measurements. You analyze what you can, and lose the tracks at the very edges of the sounds. These may not be of interest, depending on how the segmentation was done.
-
Pad sounds with silence after extraction. In some cases users may want measurements right up to the edges of the sound, but may not want to pad with the original context. In these cases, Fast Track includes a tool that allows for a specified amount of silence to be added to a set of sound files in batch. This has the downside of providing an analysis on stimuli that have been modified from their original form.