Video tracking of mouth shape was used for the composition Alignment. From the project report:

Video processing was done with EyesWeb 3.3 running on Windows XP. The patch captured the “blob” of the performer’s mouth and drew a rectangle around it. The height of the rectangle was mapped to delay feedback amount (between 0.0-1.0), and area affected delay resolution. The relationship between mouth height/delay feedback was linear, but the mouth area/delay resolution mapping scheme was logarithmic, so that opening the mouth wide could produce long delays, but the miniscule resolutions (between 0 and 100 ms) could be easily accessed with subtle movements – movements about as subtle as those required to whistle different pitches. We discovered later, while experimenting with the instrument, that the performer can make these comb filtering effects quite loud by opening the mouth wide (increasing feedback to maximum), while sticking out her tongue (!) which thereby decreased the area (resolution). And so “tongue delay” was discovered and promptly written into the piece.

eyesweb patch digram