Noise Floor and Dither

If a recorded source has a noise floor of -70db, is there a still a benefit to dithering output when saving 32 bit as 16 or 24 bit?

Dither is used to mitigate “quantization noise”.

When converting to 24-bit, both quantization noise and dither noise are both at such a low level, that for practical (listening) purposes, it doesn’t make any noticeable difference whether you dither or not.

When converting to 16-bit, yes it is recommended to use dither, even if the noise floor is the same or a bit higher than the level of dither noise. The benefit of dither diminishes as the noise floor increases, but dither still has some benefit up to the point where it is drowned out by the noise floor. When dither is completely drowned out by the noise floor, then it doesn’t matter if you use dither or not.

Generally I would recommend leaving dither enabled (“shaped”), and always ensure that your Audacity tracks say “32-bit float” (in the panel on the left end of the track).

Note that if you apply a fade to a recording, then the noise floor also fades, and that is when quantization noise becomes most noticeable, unless you use dither.

Thanks Steve.

What noise floor level would likely drown out the dither? -65, -60, -55, …

Looking at buying a 24 bit device. So probably won’t matter anyway. Still though want to know what conditions benefit or don’t.

That depends on the kind of noise. White noise will drown out dither pretty quickly, but a low frequency “hum” type noise would probably never drown it out (dither would still be required to avoid quantization noise). As I hinted at in my previous post, I’d recommend leaving the setting at default unless you have a specific (and valid) reason for changing it.

Whenever people worry about this I say, “Do whatever sounds best to you!” :wink:

But at 16-bits of better, you can’t hear dither or the effects of dither under normal listening conditions so it’s one of the last things I worry about.

You might hear it if you crank-up the volume during a fade-out. And that would generally be a fade-out that you apply, not an existing program fade-out where the existing noise doesn’t fade-out.

If you crank-up the volume that much during the whole song/program you’ll probably be listening to distortion, or if you have a system (or headphones) that gos that loud without distortion you’ll probably experience a temporary threshold shift (temporary deafness) so you won’t hear it.

The “general rule” is: Dither whenever you down-sample. If you already have 70dB of noise, a tiny bit more noise (dither) probably won’t be audible under any conditions.