All About Dithering

I remember reading a PDF by Izotope on dithering. I did a few of the tests they suggested and it seemed both the amount and type of dither did affect the lower sound levels to a small degree. I mean, I wouldn't buy a program to get a hold of the different types of dither but I would say it is nonetheless good to have. I mean, at -90dB dither isn't heard and as little of a difference as it makes I say the benefit outweighs the drawback. Dither away!
 
Oh yeah another thing I should add, (that I think Moses said basically already), if you do decide to dither, the shape/type of it is basically a bunch of bullspit designed to get you to buy some "cutting edge" program. I've heard some people swear noise shaping of dither actually plays back worse on some systems where the live mixer brightens the mix. I honestly can't tell. Like I said before, -90dB is pretty f****** hard to hear.
 
Hands up who can really hear dither?!

You! ...as soon you learn what to "look" for. Nobody really perceives "dither", but you can hear it when it's not been applied properly.


What you can perfectly hear is Quantization Distortion. It happens as soon you reduce the precision of a discrete signal (which means "digital" or "stepped"). For example from 24bit to 16bit, or when an analogue signal hits an AD converter.

Such a kind of conversion is simply done by "throwing" away the lowest bits, effectively reducing the amount of decimal places used to describe each sample. As mentioned above, we're talking about really low volume signals, the lowest your bits can carry. And here's the nasty part: Such a truncation introduces a very ugly type of distortion, one that adds very non-musical and easily perceivable periodic garbage. This kind of distortion is nearly impossible to remove and it builds up with each conversion. The latter happens much more often than you may think: For example, most of todays plugins use 64bit floating point precision for their internal calculations - but nearly all DAWs are still running with a 32bit floating point precision. It's not unusual to see 50 plugins on a mix, you'll have at least one truncation step per plugin - the added distortion will be low for each instance, but the distortion pattern will always look exactly the same and will sum up. In theory. Thank god there's a cure for it.

There's a very easy method to reduce the patterns/images created during truncation. All you have to do is to add a very small random signal ("noise") to the file BEFORE truncation. This breaks the periodic nature of these patterns and basically "exchanges" them with noise. This small random signal is called "Dither". It's works nearly perfect if done right (so much that you can hardly hear the difference between the 24bit original and the 16bit CD version).


Ok, I still haven't answered your question. ;) Dithering shouldn't be audible, so it's a good idea to get familiar with the bad side. Do a simple experiment:

1. Take a 16bit or 24bit file with a huge fade-in or fade-out.
2. Save a copy in 8bit, DO NOT USE any bit-depth converters, just keep it raw.
3. Listen! Compare the original with the low-bit version, closely listen to what happens as the fade attenuation gets stronger..

Now, do the same thing as before, but add an 8bit dithering to the original before you save the file in 8bit.

Compare both 8bit versions. Now don't tell me you can't hear the effect of dither.
 
Last edited:
Bob Katz once wrote

"[quantization distortion] can introduce harmonics, subharmonics, aliased harmonics, intermodulation, or any of a set of highly undesirable kinds of distortion. In order to prevent this, the signal is dithered, a process that mathematically removes the harmonics or other highly undesirable distortions entirely and replaces it with a fixed noise level... dither let's us encode low level signals below the -96dB limit... adding noise is not dither. Dither's resolution is truly physical/mathematical in nature, not merely a trick which fools the ear. Dither is not simply a means to mask the low level digital breakup... we can hear signals below the noise which explains why the perceived range of the dithered system is greater than its codability."

Hope that helps. I hope I didn't overuse the ellipses. It was basically a bunch of psycho acoustic, technical jargon or just redundancies.
 
Who cares about dither?
I do ^_^

Someone made a good point about it being of little use at high resolution, but is very usefull at what it does.
You could say at high resolution its effects are neglegable but to some extent the same can be said if uhf/sub harmonics, we dont actually hear them as such but something has happened to the sound.

I made a comparrison a few years back and i did notice a difference in the sound,
I suppose if it has to be used its best to find out what you preffer rather than just a random choice.
 
I always use dither, never understood it though...just used either 24bit (for normal audio) or 16 for CD press. just becasue i was once told that.. but bever botherd to research it.. so glad for the real info now.
 
I always use dithering maybe cause I can't tell much difference in sound. Maybe I need a better Sound Card?
That shouldn't do it. Unless for some odd reason your current sound card doesn't play anything lower than around -90 dB. The very nature of dither is so that it is NOT heard. If your really want to hear dither though try listening to 8 bit, highly noise shaped dither and you should easily hear a piercing HF noise.

But when using dither you shouldn't hear anything except the lack of quantization distortion.
 
Dithering is the most common means of reducing the color range of images down to the 256 colors seen in 8-bit GIF images. Dithering is the process of juxtaposing pixels of two colors to create the illusion that a third color is present. The more dither patterns that a device or program supports, the more shades of gray it can represent. Thanks.:pointing:
 
Back
Top