Speech turn, gaze, and blinking are all annotated using Elan and Praat together. Good hardware is necessary to allow smooth frame-by-frame replay of two parallel videos in Elan. I have found recent Apple laptops to be well up to the job, but have had poor results using various flavours of Windows.
Modus Operandi
To annotate a conversation, we start with two video files (.avi) from the IFA website, and the single large sound file (.wav). These three are loaded into Elan (screenshot). We also read in the sound file into Praat, taking care to read it in as a "long sound file", so it does not hog memory. In addition, an annotation tier is created with six tiers. For speakers A and B, these would be called Ablinks, Agaze, Aspeech, Bblinks, Bgaze, and Bspeech. The blink tiers are point tiers, while the other four tiers are interval tiers (screenshot). See Praat documentation if that seems opaque.
We work with ten seconds of data at a time. First select the ten seconds in both Elan and Praat. I have found it convenient to first listen to the selected stretch in Praat, and to annotate Speech Turn for both subjects. Then I step through the video, frame by frame, annotating blinks and gaze for the first subject, then I repeat this for the second subject.
In Elan, it is possible to scroll slowly back and forth over an event of interest (on the Mac, this is Command-right/left arrow). This is very useful in the disambiguation of complex eye movement.
Measurement Accuracy
The video is recorded at 25 frames/sec, which provides an absolute limit to temporal accuracy, such that events separated by less than 40 ms can not be disambiguated. In practice, accuracy for registering blinks is probably more realistically estimated to be about 100 ms. Details of blink morphology and means for maximizing precision are provided in the Blinking page. Gaze shifts can be captured with equal precision. Speech Turn, as documented elsewhere, is harder to define with precision and will inevitably entail some subjective estimation on the part of the annotater.
A Caveat When Using Praat
Familiarity with Praat is assumed here. The programme comes with extensive help and documentation and is wiedly used. One thing to watch out for in the present instance is the fact that stereo recordings are made using two separate microphones. This provides a rough channel separation, but there is bleed from one channel to the other. Particularly when one speaker is silent, it may take care to work out which channel is being talked on, and to avoid confusing speech of one subject with speech of the other.