Now, alter the distribution significantly, so that the sum of each sound's array is roughly the same, but the tops and troughs swap place. After this, recalculate the sum for the frequency for each of the columns (corresponding to frequencies at a position), so they add up to one. (Also, try maintaining the sum of all vowels vs. the sum of all consonants in a given column somewhat intact.)
The result would be a language whose sounds are the same as English, but the way they are used is significantly off. I do think using a restrictive and relatively English-like set of consonant clusters might make sense, but who knows.
Usage for this thing: naming languages in fiction.