Why should I dither?

[MichaelAlan]

Hmmm?

[nopenopenope]

dither adds random noise to the track, giving the digital audio a better chance of being sampled versus not at all. this is, of course, at super low levels. maybe with the loudness wars, dither is a bad thing. that random noise will sometimes steal .2 dB or so from you - and if you're into hyper compression, that's .2dB RMS you're missing out on!

but in all seriousness, in my experience it helps create cleaner attacks and actually lowers the perceptible noise floor... assuming you're listening to a recording that utilizes the full dynamic range of a given medium.

to each his own.

[leigh]

dither adds random noise to the track, giving the digital audio a better chance of being sampled versus not at all.

Sorry Josh, no.

Dither adds noise to the track in order to mask quantization noise.

Do you like quantization noise? Then don't use dither.

In order to hear what quantization noise sounds like, put a bit crusher plug-in on a track. Start reducing the bit depth... by the time you get down to around 6 bits it should be quite audible on any source material.

Of course, bit reduction to 16 bits causes quantization noise at a much quieter level than at, say, 6 bits. However, the quality of that noise is the same - an inharmonic overtone, with an audible effect of "grittiness" or "edginess".

See also: http://en.wikipedia.org/wiki/Dither

cheers,
Leigh

[Jeff White]

Download and read this PDF: http://www.izotope.com/products/audio/ozone/guides.html

[MichaelAlan]

Thank you. I am now schooled. Could this account for some of the harshness in my recordings? I try to get the highest clarity and best signal to noise ratio possible, but I still end up with an edginess... Could this also be just a result of having no outboard/analog gear...?

[David Piper]

In my experience, dither is way, way low on the list of things that are going to affect the quality of your recording. Leave it to the mastering guys. And as I understand it, dither is mostly important to very quiet signals, near the noise floor. Like reverb tails or fade-outs. I can't hear the difference between a dithered and un-dithered signal in the middle of a song. Don't worry about it too much.

[nopenopenope]

Dither adds noise to the track in order to mask quantization noise.

I agree. The random noise gives signals closer to the 0 crossing point a better chance at being sampled versus being ignored. The theory is "something is better than nothing", and according to my ears, I'd be inclined to agree. Perhaps I was too quick to say simply 'super low levels', as that's not a totally accurate way to describe it outright.

try recording something at 24bits, then bounce to 16bit without dither and at a very low level. gain it back up - you'll hear it!

[leigh]

The random noise gives signals closer to the 0 crossing point a better chance at being sampled versus being ignored. The theory is "something is better than nothing", and according to my ears, I'd be inclined to agree.

Your explanation is not consistent with how I understand digital audio. A signal is always "being sampled", whether the value of that signal is 0 or something else. So adding noise can't change this.

Even if I changed your wording around to something like "The random noise gives signals closer to the 0 crossing point a better chance at having a non-zero value", that's not true either... noise can subtract as well as add amplitude, so it could take a signal that was non-zero and make it zero.

The theory of dither is to hide the quantization error that occurs when reducing bit depth, by covering it over with noise. There's plenty to read out there about how this works, but the bottom line tradeoff is one of replacing signal distortion with broadband noise.

Not trying to pick on you, just wanting to clarify the whole process.

[ashcat_lt]

The Ozone guide linked above is a good read on this subject, and it has pictures. From what I gathered from that, and other sources I don't recall right now, dither actually reduces quantization distortion. It might mask some as well, but it sure looks like adding dither makes the distortion artifacts themselves disappear, at least partially.

[leigh]

The Ozone guide linked above is a good read on this subject, and it has pictures. From what I gathered from that, and other sources I don't recall right now, dither actually reduces quantization distortion. It might mask some as well, but it sure looks like adding dither makes the distortion artifacts themselves disappear, at least partially.

I just downloaded that "Dithering with Ozone" guide, and despite its specific-sounding title, it really does have some great general explanations of how dither works. Well-written!

The difference between "masking" and "reducing" quantization noise in this case seems to me to be one of semantics. Are you actually reducing the quantization error? No, the only way to do that is to use more bits. So, instead of letting this weird sonic by-product of bit reduction rule the LSB, you're injecting noise, so that the LSB moves in a way *unrelated* to the the greater waveform. (Without dither, the LSB is modulating in a way that is related to the greater waveform - and hence it is a form of signal distortion.)

Anyways, I hope that the Dithering with Ozone guide gets referenced more often around here... it provides a nice intro to the topic.

cheers,
Leigh