I wonder what the concensus is on dither during silence periods, considering that TPDF raises the noise floor, and Noise shaped applied dither even more, so what if a piece of music has 1 second of silence (pure 0), apply dither or not to that silence section? would the ear be more likely to detect the dither when it kicks in again?

I've heard some tunes (namely in daft punk's Discovery) that only dither when no digital silence is heard, and the feeling is quite annoying - It sort of sounds as if your player skipped... I think it's a better idea to dither everywhere or no where.

It's pretty unusual for dither of any normal level at 16 bits to be audible in a normal playback gain structure.

If you do turn dither off, you'll basically be inserting noise modulation.

If you do turn dither off, you also create the following problem:

When does the signal "start"? If you can't define that, then you can't turn dither off.

If you can define that, why are you writing zeros to the media?


Basically, dither should be constant. If you don't have constant dither, you really aren't quantizing right, even if you're quantizing zeros.

Oh, and why does noise-shaped dither have to raise the total dither level?

>Oh, and why does noise-shaped dither have to raise the total dither level?

I was a little unclear, the noise from noise shaped dither is mainly in the upper frequencies using the area above audibility (on over generalization as there are tens of differing noise shaping methods all with differing frequency curves, mainly to increase the range at frequencies which count, sacrificing lesser important frequencies) at a higher level than say the levels from rectangular dither (which is spread across all frequencies).

I'd say it depends on how (with what kind of weighting) you measure the "total level"

BTW: The quantization noise is supposed to be independent from your signal (steady loudness of noise). That's why we use dithering! -- even on silence

Sebi