MOTM

One easy way to extract voiced/unvoiced information (as I recall) is 
to use a high-pass filter with a cutoff at 12 kHz or so.  Plosive and 
fricative sounds have more energy in the higher regions of the 
spectrum, so when an S or F or P or whatever is spoken, the HP filter 
passes the high spectra which is rectified and used to trip a Schmidt 
trigger (threshold detector).  This is used to control an electronic 
switch, which switches from the default carrier to a noise source.  
As soon as the unvoiced sound is over, the switch reverts back.

johnm