QUOTE(Frank Klemm @ Feb 15 2003 - 10:36 PM)
This answer is completely wrong for WinAMP 2.x .
I hope Garf can explain why.
You are right - sorry.
WinAMP 2.x does
compressed data -> decoder (32 bit float) -> 16 bit -> DSP -> output
You want the dithering on the 32 bit float to 16 bit conversion, so at least in the decoder plugin. I'm not sure about dithering multiple times in this setup, logically it would be needed for each plugin that does 32 bit -> 16 bit conversion internally, but I do not know how noise adds up. I think noise would be actually minimized if we dithered each time, is this correct?
For FB2K my advice holds, because it does
compressed data -> decoder (32 bit float) -> 32 bit -> DSP -> 32 to 16 bit conversion+(dither) ->output