You can specify the amplitude of the noise either as a percentage of the maximum amplitude or as an exact number of points. For example, 10% is roughly the same as 12 points in a 8-bit sound, or 3276 in a 16-bit sound.
Dithering can be very useful when downsampling a sound to a lower sample size. By just stripping the last few bits of the samples, the original waveform is strongly distorted. If the sound is dithered before downsampling, a little noise will be added, but that is far more desirable than distortion.
