I have no idea about how these programs function. Do they concentrate as much on what is to be kept? If the voice and dialog is of greater volume across the same frequency as the lower volume sound being eliminated, do these programs have a way to extrapolate from the point where the frequencies begin and end their intersections to "fill in" the desired sound and identify the undesired useing volume differences? Can they take quick samples of what's happenning just before and just after intersection to make guesstimates? This seems like the sort of problem where the 2of3, 3of5, 4of7... method for validating data streams might have application. If fuzzy-logic sort of processed samples of what is to be kept are created 5 different times and samples of what is to be eliminated are created, let's say 3 different times, then the most in common 3 of the 5 kept sound samples and 2 of the 3 most in common samples for sounds to be eliminated could be selected. Then the 3 kept and 2 elimination streams could be compared together to determine the most in common 3 out of the five. Like I said, I don't know how the commercial software operates. :)