Music Technology 101: Dithering Explained (2/2) - What, Why and When to Dither

In this two-part video tutorial I will explain dithering from the ground up. For your convenience, here are the links to the two parts:
Part 1: • Music Technology 101: ...
Part 2: • Music Technology 101: ...
You do not need any special background in signal processing, audio or dithering to follow the current videos. However, you should know what bit depth means. If you don't, fear not! Just watch my short video tutorial about bit-depth and sampling rates right here:
• Music Technology 101: ...
What's in Part 1: Dithering is all about getting rid of quantization noise. What is quantization noise? Glad you've asked, because that's exactly what we're going to cover in the first part! Shortly put, quantization noise is the noise introduced whenever we reduce the bit depth of our signal. For example, most audio is recorded using 24 bits of resolution, but modern audio CDs only have 16 bits of resolution, implying that a reduction in bit depth must be applied. This reduction will introduce some artifacts known as quantization error, or quantization noise. This "noise" will have some jarring, unpleasant frequency components which we'd like to get rid of.
What's in Part 2: In the second part we will cover dithering. To "dither" a signal means to add some form of random noise to it because lowering its bit depth. This dither noise has a beneficial effect: while it doesn't eliminate quantization noise, it gives it a more random, "white" nature which is less disturbing to the listener. When the amount of white noise equals approximately 1 bit in magnitude, the quantization error becomes a lot like white noise. This is because quantization involves rounding the input signal either up or down. When the noise becomes on the order of 1 bit, the rounding becomes random, and therefore the quantization error becomes random as well.
DITHERING TYPES
Dithering requires that we add random noise to our signal before downsampling. This noise should have a flat spectrum - in other words, be white. However, there is more than one way to generate white random noise. Probably the easiest and most efficient way is to use what's known as a triangular probability distribution function, or TPDF. You might have seen these initials in your dithering plugin. This is an excellent way to efficiently dither. Although we won't discuss the heavy mathematical theory of dithering in this video, I'll just mention that TPDF white noise decouples the first and second moments of the quantization noise.
Another option consists of using noise with a non-flat spectrum. For example, you'd might add noise that has more high-frequency components, such as Blue Noise. This is referred to as shaped noise, shaped noise dithering, or colored dithering. What this tries to do is force the dithered quantization noise to occupy higher frequencies that are outside the human audio range. Once again, personal experimentation is key to deciding whether you want to use colored dithering or not, but this is truly a very fine point. You will be fine if you just stick to TPDF. However, a word of caution: only apply colored dithering at the FINAL stage of your processing. If you need to dither audio at some point DURING mixing, use TPDF. This is because subsequent processing of the audio can cause the colored noise to creep into the audible listening range and create nasty artifacts. So: Use TPDF at all stages before mixing, and use TPDF or colored dithering during the final mixdown.
LINKS OF INTEREST
Here is a wonderful guide to dithering written in 2002 by Nika Aldrich, targeted at the audio engineer:
www.users.qwest...
This is truly geared towards the audio enthusiast and does not go into any math. It is heavily illustrated and references industry standard plugins such as Apogee's UV22.
Wikipedia's entry on dithering: en.wikipedia.or...
My Other Videos
My KZbin channel has many other video tutorials covering various topics in both audio and music, mostly geared towards piano playing. Here are a few examples:
Bit depth and sample rate explained: • Music Technology 101: ...
Song writing Tips and Tricks - Rhythmic Doubling: • Free Songwriting Secre...
Reading Sheet Music for Beginners: • Reading Sheet Music fo...
The 2-5-1 Harmonic Progression Tutorial: • Harmony 101: The 2-5-1...
Playing Left hand Piano Arpeggios: • Best Free Piano Lesson...
An Exercise for Developing Piano Right-Hand Technique: • Developing Piano Techn...

[@rcarrillo7]

Wow, these are so well-produced that it makes it clear as day. Thank you. I subscribed.

[@MangoldProject]

+Ruben Carrillo Thanks Ruben! I hope you find more stuff to your liking. I do mainly playing techniques/ideas videos, but I've been contemplating doing more music technology style videos like this one. At any rate I appreciate the feedback.

[@alisaljic]

For our convenience, could you kindly make a playlist with all three Music Technology 101 videos. And, please, do make more videos explaining the rest of the nuts and bolts of music technology. You sir have a praiseworthy way of conveying knowledge. Thank you for taking the time to do all this.

[@rorimckenzie237]

That was fascinating! What I learned here probably won't be useful for the higher music tech exam, but I'm very interested... thank you!

[@Alienadin]

Great video. Very well explained. Thanks.

[@EliseBoulanger]

I just watched your three videos on sampling rate, bit depth, and dithering. They were super helpful, clear, and effective. Thank you very much! I hope you make a video on noise shaping. I am training to use the TASCAM DP-24 Digitial Portastudio. This unit has a 'noise shaped dithering' feature, which makes me confused as to why they use this label after what you said.

[@anteneharayaselassiealula7184]

I have never commented on youtube ... you are amazing thank you

[@MangoldProject]

Thank you! I'm glad I could help you out.

[@compiutershick1]

Thank you for the video! It's very helpful.

[@Dirk___]

Great video! If you have time, do a video about noise shaping. It will be good.

[@MangoldProject]

Thanks! Time (or lack of it, to be precise) is the key word here,:). I'll do my best when I get some time off.

[@josephjwoods66]

I 100% got it now, thanks for making the outstanding video.

[@AA-lq5pu]

Brilliant!

[@milenadeltorto7158]

Thank you!

[@84nomas]

good.really good.keep it up man, keep em commin

[@MangoldProject]

+84nomas Workin' on it :)

Great videos, thank you very much.

[@MangoldProject]

+Ahmet Gökhan Coşkun You're welcome Ahmet, and thanks for watching!

[@TheObligation]

Very interesting. Could you make one of these about interpolation?

[@MangoldProject]

Calixte De Coster I think interpolation would make for a highly technical video which would really fit more on an engineering channel. Plus there are some good ones on KZbin already, so I don't think I can improve on them.

[@Riberablues]

Great. Nice thinks are simple thinks.

[@roderik1990]

A note here, floyd steinberg is more akin to noise shaping than it is to dithering. Despite it being called dithering itself.

[@Chillos100]

best teacher! thnxx..

[@ricoconsulting]

very very good well done

[@badhalf7024]

Great videos.

[@MangoldProject]

Why, thank you, that is very nice of you to say :).

[@refaelshansi]

Hi, Thanks for the video. I have a question regrading the noise that you add in the dithering: should the noise level PSD should be limited by the value 0.5X(q/srt(12))^2? Since for my understanding, when adding to much noise, higher than ( LSB/2), you can degrade the ADC performance. Will appreciate your explanation. Refael

[@MangoldProject]

I don't really know what the noise's optimal pdf (or psd) should be - that's a deep theoretical question that should be addressed to a signal processing expert who specializes in dithering.

[@MangoldProject]

Thanks. I do my best to simplify :)

[@symoundreddurias6402]

You should be a teacher. Much love!

[@MangoldProject]

+Symoun Dred Durias I am! But not music (I'm a scientist and also occasionally teach at my research institute :) ).

[@mika461983]

should I engage dither at the start of mixing the project? The sound from DAW came out with 32 bit float format (or higher) & goes to DAC, which is 24 bit, so all we hear is truncated?

[@jasonhoumusic]

The 32bit floating point is used by DAW's summing engine. The use of floating point is just for internal calculation purpose. So no need to worry.

[@mika461983]

@@jasonhoumusic so, how many bit are going from DAW to DAC actually? 24? if so, then it means that there's some default dither process on the DAW's output section?

[@Plaswuff]

Thanks man.

[@MangoldProject]

You're welcome!

[@codemiesterbeats]

thanks buddy

[@GHovahBeats]

I saw a video by a guy named Monty from xiph.org about stairsteps in digital audio. He said that stairsteps simply doesn't exist and that the output signal is identical to the input. Wouldn't that mean that quantization errors doesn't exist either since they are caused by stairsteps? Im a bit confused lol. Other than that great video!

[@matiaskok2457]

+G-Hovah Beats If I understood this correctly, the quantization gets introduced to the audio if you convert it from a higher bit depth to a lower bit depth. I'm still learning this stuff as well so I could be very wrong. lol.

[@hendrix5757]

Quantization errors will occur and be present in any and all digitized audio, regardless of bit depth. However, the quantization noise's amplitude can be minimized by using larger bit depths and dithering when lowering bit depth. The 'stair steps' are simply a visual/graphical representation of a digitized waveform, akin to how a pixel is represented as a basic square on screen, but does not affect the fidelity in how the waveform is reproduced. The samples themselves only store values for specific amplitudes over time, but do not contain any additional data in between each sample point.

[@enricocasonato2222]

Yep stairstep wave is not "real" let's say because the digital to analog converter produces a sinewave, but that sinewave will only approximate more and more precisely the higher the bitrate goes but still with some error, that produces that noise. It's just the graphical representation you see that is wrong not the concept

[@roygertel]

Hi ! I have questions that I didn't get straight /consistent answers to, maybe U can: I'm working inside Cubase 9 with 48 khz / 32 bit float resolution. Can I export the final file in the same resolution (without dithering, no changes at all) & then upload it to KZbin ? So... KZbin will make the first & last compression for me, from the best resolution I uploaded the file. Does it will make artifacts to the file ? because, in that scenario the compression will be directly from my 48 khz / 32 bit float resolution to KZbin AAC (or MP3) format ? Does the file actually needs dithering for that ? Do U have idea what will actually be the best workflow to send KZbin the exported file ?

[@MangoldProject]

KZbin audio is too crappy for this to matter.

[@debrucey]

quantisation noise is completly inaudible for 16 bit audio at normal listening volumes, don't worry about it

[@tranlequocdat]

i'm sorry if my question is stupid :v how can i get the random noise? for example If i turn a song as 24bit to 16bit, we have to add random noise right? but how we get the random noise and how to add it? Can you describe more clearly about this process? Thank you!

[@MangoldProject]

Trần Đạt Hi Tran. You're not supposed to convert 24 bits to 16 bits by "yourself" - that's what software plugins are for. Whenever you select downconversion in Cubase/Protools/whatever the software will already automatically inject the small amount of (pretty much inaudible) white noise needed for dithering. So no need to worry :).

[@tranlequocdat]

MangoldProject Thanks for rep! ^_^

[@maltebergman5242]

Nice vids, I'm a hobby mixer and I had no clue what UV22HR in cubase did until I googled around and found your and a few others' videos. Hopefully it will help a bit (haha nerd humor!) in the mastering process for me =) Thanks a lot for the help!

[@MangoldProject]

+Malte Bergman Just saw your comment - must've missed it earlier on. I support your nerd humor and hope I've helped you master the concept (see what I did there? :) ).

[@maltebergman5242]

I see man! ^^ yeah it helped a lot, I had "mixed" thoughts about what the plug even did so I went searching and here I am, doing lame sound geek jokes :S

[@MangoldProject]

+Malte Bergman I enjoy musical jokes. Here's one of my favorites: a drummer walks into a musical instruments store and says, "Everybody makes jokes of how dumb drummers are. I'm fed up with it and have decided to switch! Give me that red trumpet on the wall and that white accordion next to you on the floor!" The seller looks at him and says, "Listen, you can have the fire extinguisher, but it's pretty cold here and I really need my radiator ... " (no disrespect to drummers, I could never compete with someone who has to synchronize four limbs! :) )

[@maltebergman5242]

xDD

[@MrSoulja205]

I have fl studio 11 do dither when I export?

[@AlexTheAverageKid]

+Cory Mckenzie I know this comment was from over a year ago - but from what I can tell, don't dither unless you are downsampling or have a track with lots of dynamic range. I use FL11 too :)

[@NotOrdinaryInGames]

Now I need someone to explain to me the rectangular kind.

[@NotOrdinaryInGames]

Editing becomes even EASIER once you are able to work in 32 bit.

[@davidbridgeman]

Actually dither replaces quantization noise it doesnt mask it.

[@MangoldProject]

As I state in the video, dither IS noise that is added to a signal to help make quantization noise assume a random character. It does not replace it, and it only "masks" it inasmuch as it turns it from having some recognizable and jarring spectral content to having a random, white or white-ish unobtrusive spectral content.

[@davidbridgeman]

MangoldProject from xiph.org "Dither doesn't drown out or mask quantization noise, it replaces it with noise characteristics of our choosing that aren't influenced by the input." From wikipedia " a process that mathematically removes the harmonics or other highly undesirable distortions entirely, and that replaces it with a constant, fixed noise level"

[@MangoldProject]

delta214 Re-read my answer: "it turns it [quantization noise] from having some recognizable ... spectral content to having a random ... spectral content." This is exactly what your quotes say: "a process that ... removes the harmonics or other ... distortions ... ". Both say the exact same thing: dither changes the quantization noise's characteristics (or shapes its probability density function, if you want to be precise). If you want to use the word "replace" instead of "change" or "turn" then by all means, go ahead.

[@davidbridgeman]

MangoldProject alright. Don't want to argue or anything. Just getting it clear in my head. Thanks for the video :)

[@venkateshreddy1044]

Complicated Tutorial

[@MangoldProject]

Complicated reality