July 22, 2025 • 47 mins
Ever wondered why engineers intentionally add noise to audio? In this episode of Inside the Mix, returning guest, mastering engineer and Berklee professor Ian Stewart, breaks down what dither is, why it’s used, and when to apply it in your music production workflow.
Through clear explanations and real audio demos, Ian demonstrates how dither replaces digital artefacts with smooth, barely noticeable noise, preserving details such as reverb tails and low-level audio that would otherwise be distorted during bit depth reduction.
You’ll learn:
- Do I need to dither when exporting a 24-bit file to a 16-bit file?
- Can skipping dither cause distortion in quiet passages?
- When should you use dither in mastering?
Whether you’re mixing in 32-bit float or exporting for streaming, this episode explains how audio dither helps preserve quality during final delivery and when it’s best to skip it.
Links mentioned in this episode:
Follow Ian Stewart
Ways to connect with Marc:
Book your FREE Music Breakthrough Strategy Call
Radio-ready mixes start here - get the FREE weekly tips
Grab exclusive access to BONUS content
Follow Marc's Socials:
Instagram | YouTube | Synth Music Mastering
Thanks for listening!!
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Marc Matthews (00:00):
Ever hear the term dither in audio production
and wonder do I actually needthis, or is it just another myth
?
I think myth is probably thewrong word there, because it's
not a myth, it does exist.
What if I told you that addingnoise to your final bounce could
actually improve the way yourmusic sounds?
Well, that's exactly what we'rediving into today.
Ian Stewart (00:22):
You're listening to the Inside the Mix podcast with
your host, Mark.
Marc Matthews (00:26):
Matthews, welcome to Inside the Mix, your go-to
podcast for music creation andproduction.
Whether you're crafting yourfirst track or refining your
mixing skills, join me each weekfor expert interviews,
practical tutorials and insightsto help you level up your music
and smash it in the musicindustry.
Let's dive in.
(00:46):
Hey folks, welcome to Insidethe Mix.
This is part two of a specialseries with mastering engineer,
educator and plugin developer,ian Stewart.
In part one we explored samplerates and busted some big myths
around digital audio.
If you haven't heard that oneyet, be sure to check out
episode 203.
And today we're zooming in on atopic that's even more, I say,
(01:10):
misunderstood.
I guess it could be wellmisunderstood, and that is
dither.
So a bit about ian, if this isyour first foray into inside the
mix.
He's worked with um artistssuch as krs1 and countless indie
artists.
He co-developed a baseline proand teaches mastering at
Berkeley as well.
So you know you're learningfrom someone who lives and
(01:30):
breathes mastering.
He lives and breathes thisstuff.
Ian, welcome back.
I'm going to say how are you?
But we're recording theseepisodes and series at the same
time, but I'll ask it anyway.
How are you?
I'm great.
Ian Stewart (01:43):
As you can see, I changed between it.
No, um, yeah, no, it's, it's.
Marc Matthews (01:53):
I'm actually getting a little bit of a breeze
through the window here, whichis really, really lovely on this
balmy new england day.
Yeah, I am jealous because inin my studio here I don't have
uh windows.
Ian Stewart (02:00):
It is essentially no windows in a mastering room
are weird, but that's whathappens when you turn a guest
room in your house into yourmastering room yes, yeah, yeah,
I.
Marc Matthews (02:09):
When it gets to the summer, I'd be thankful.
Instead, I've got a a a heaterthat has a cooling feature on it
, so I'm having to use that.
It's pretty grim, it's prettygrim and it gets hot in here
towards the end of the day.
Ian Stewart (02:19):
Yeah, no, once I have the, the tv and all the
analog gear switched on, it'sjust yeah, it gets absolutely
brutal in here.
Marc Matthews (02:26):
I bet man.
So this episode is going to beyour go-to guide for
understanding what dither is,how it works and why it matters.
So we're going to be talkingabout when to apply it, what
happens if you skip it andwhether it makes an audible
difference in real-worldscenarios.
So if you've ever explored yourmix wondering should I hit the
dither button or not, this oneis for you.
(02:47):
But before we dive into that,if you want weekly tools and
tips to your inbox to make yourmusic playlist worthy, click the
link in the episode descriptionand get my free weekly tips to
elevate your music production.
I say my weekly tips not justmine, but I also get guest
contributors as well.
No fluff, no spams.
Just once a week you'll getthose to your inbox.
So in when we're talking aboutdither, I think, as we did with
(03:10):
sample right, it's probably agood place to start.
What is dither and why is itused?
Why do you?
Ian Stewart (03:15):
yeah, so most simply I mean you kind of said
it in your in your intro theredither is noise.
It is just broadband noise,usually kind of white noise,
although we can shape it.
We can use something callednoise shaping.
We'll get more into that, I'msure.
But it's very low level, veryquiet, and it is a signal that
(03:40):
you add in when you are, signalthat you add in when you are
technically, when you arequantizing audio.
So I'm going to refer back tolast week slash 40 minutes ago.
Yeah, yeah, time is an illusion, lunchtime doubly so.
In the words of douglas adams,I think.
Um, so we talked about takingsamples 44,100 times or 48,000
(04:06):
times a second right.
So we can do that.
We can slice up the, the audioas it's coming in from our
microphone through our mic preinto our converter.
We can slice it up and then wehave to measure the voltage, and
that part is all pretty muchcompletely.
There's no loss there.
Right, the time slices are whatthey are.
(04:29):
You don't have to compromise onthem.
You can kind of make themwhatever rate you want.
There's just a clock that runsthem very precisely.
But then when you measure thevoltage, you have to turn that
into a digital value and youdon't have infinite numerical
values that you can store indigital.
And so the more kind ofresolution or the bigger a
(04:50):
number you want to be able tostore in digital, the more bits
you need to use.
So one bit can represent twovalues, zero, or one.
Two bits can represent fourvalues, zero, one, two, 3, 4.
It's kind of an exponential-ishthing.
We won't go too deep into that,right, but more bits equals
(05:10):
more values.
So at 16-bit, which was the kindof standard for CD, I say was,
it still is, cds are and everwill be 44.1 kilohertz, 16-bit.
But we don't use CDs as muchthese days.
I'm sure some people still do,but they're they're hardly as
common.
Um, 16 bit gives us enoughvalues that the kind of residual
(05:37):
noise that's left over and weagain we can unpack more of this
.
I'm trying to keep it highlevel at first and we can go
deeper, of course.
Yeah, yeah, um, the residualnoise from kind of having to
round your signal up or down tothe closest level is pretty low.
It's down around 96 DB belowpeak level, right, your clipping
(05:59):
point.
And that's, if you've everheard that the dynamic range of
16-bit audio or CD audio is 96dB of dynamic range.
That's where that comes from.
But basically, right, when youmeasure that voltage, it may
land right at a sample value,right at one of those digital
numbers that you can store, butit could be anywhere in between.
(06:19):
And if it's in between, youhave to either round it up or
down.
You round it to the closestvalue.
And when you do that you haveto either round it up or down
you round it to the closestvalue.
And when you do that, you addthere's a little distortion,
there's a little leveldistortion.
That's being added because it'snot exactly right.
Right, the level that you'restoring it at isn't exactly
right, and a lot of people thinkof this as like graininess or
(06:47):
something like that.
Really, what it does is itdetermines the level of noise,
how quiet this like white noisesignal is below the peak level,
and part of that just like lastweek we talked about aliasing
filters right, this low passfilter that prevents high
frequencies from getting in,being built into your converter.
It's not something you have todo, it's just built in.
One of the other things thathappens when you record a signal
.
When I plug my microphone intothe preamp and it goes into the
(07:10):
converter.
Dither is being added to that.
Now I'm recording at 24 bit sothat noise level is even lower.
It's down around negative 144dB below clipping.
But actually dither is addedwhen you record an analog signal
and that is to prevent to makethe distortion that happens be
(07:31):
noise.
It turns this distortion intonoise, it doesn't mask it.
We have some great examplesthat we'll look at and listen to
that really demonstrate this.
But we can do the same thingwhen anytime we reduce bit depth
if we're going from 32-bitfloat down to 24 or 24 down to
16, we can add dither to preventthat reduction in number of
(07:56):
steps that we can quantize tofrom being yucky distortion into
just noise.
That's kind of like tapis andso really it's noise that we add
to the signal.
That prevents another type ofdistortion from happening.
It replaces what we would calltruncation distortion with just
noise and when we have steadystate noise, especially at very
(08:19):
low levels, you kind of justdon't even notice it, whereas
truncation the distortion thatyou get from truncation can be a
little more noticeable and isnot harmonically related to the
music, much like aliasing.
There's no harmonicrelationship there, so it can
just make things sound a littleweird and grainy and gross.
Marc Matthews (08:38):
We've got some examples that we're going to go
through in this episode, whichI'm looking forward to getting
to.
One quick question 32-bit floatyeah, so DAWs now support
32-bit.
So when we're in our DAW, we'reworking away and you see a lot
of discussions again I say thisa lot on the podcast with
(08:59):
regards to DAWs and 32-bit andnot being able to clip and et
cetera when we're using them.
But could you just is there ahigh-level description you can
give of 32-bit and how it'sdifferent to sort of 24 and 16,
if that makes sense?
Ian Stewart (09:17):
Yeah, yeah.
So 16-bit.
24-bit gives you a discretenumber of values that you can.
The word we use is quantize,right, Not average around your
audio levels too, right, youhave to snap them to one of
these levels.
(09:38):
So in the example of if we havetwo bits that can store four
values 0, 1, 2, 3, 4, sorry,zero, one, two, three, really
zero through three right, thoseare four levels.
If, if a level comes in andit's at one half, we have to
make a decision whether to roundit down to one or up to two,
right, so we've got to snap itto one of these levels.
So we call that quantizing, andit's sort of like quantizing
(10:01):
MIDI, right, you're snapping itto the nearest time value.
Same idea, that's why we callit quantizing.
You're snapping it to the kindof the nearest value.
So for 16-bit, we have I thinkit's something like I'm probably
going to get the exact numberwrong, but it's something like
64,000 available values that wecan snap to.
(10:21):
That's a lot.
If that were a stack of paper,uh, it would be very tall, right
?
If each, each piece of paperwere an individual value.
It's like you could do the mathand it's crazy, right?
So already that's giving us alot of values, but it is this
defined bucket of, and and eachvalue is the same, space apart.
(10:43):
Uh, and it's.
It's like, just like if you hada ruler and each, each, each
millimeter marker, or, or 16thinch, or whatever.
You know, we're one of thesevalues, they're all equally
spaced and you have a finitenumber, floating point.
That's so, that's fixed point,that's 16 bit or 24,.
Four bit can have 32-bit fixedpoint, but it's pretty uncommon.
(11:05):
Floating point is a littledifferent.
You have, basically, the way tothink of it is you have a
decimal number and that.
So that would be like 0.
Dot, dot, dot or 1.
Dot, dot, dot or whatever, andyou only have a few digits of
(11:27):
precision there, and then youhave a multiplier, and so what
you're doing is you'remultiplying and scaling this
kind of lower precision thing upand down, and so it's sort of
like you have.
If I go back to the tape measureor ruler analogy, it's like you
(11:48):
have a shorter ruler with lessmarks on it, but you can almost
like magnify it or zoom it upand down.
I'm going to try and stay inmetric.
For most of the world.
It might be, you know, fivecentimeters down here.
(12:09):
But if you scale it way up andyou need it, you know you need
it to store a much larger value.
Now it might be a kilometerlong, but the spaces in between
those marks are also now larger.
So you could have fivecentimeters down here with
centimeter marks and then youcould go up to a kilometer, but
(12:29):
now the marks are 200 metersapart right, so it's does that
make sense?
it does, yeah, describing it itdoes.
Marc Matthews (12:39):
But in my head I'm thinking if, if that so with
regards to audio because youmentioned about the the um, the
tape, measure the ruler down, ifwe if you're watching this on
on youtube, because I'm sort ofpointing down we've got a
narrower range and then furtherup we've got a larger range.
(13:00):
So if we were to go into thatlarger range with audio, how is
that stored?
Does that not mean that we'resnapping to points further apart
?
Ian Stewart (13:12):
We are, and also the signal is a lot louder.
So the relationship right.
So if your signal goes 200 dBabove full scale, which, for
what it's worth, 32-bit float,does have a clipping point and a
noise floor, it's just that theclipping point is something
like 750 dB above zero and thenoise floor is 700, something
(13:34):
else below zero.
So it's about a 1500, 1,500decibel range that you have and
the range of sounds that humanscan hear is about 200 decibels
have and the range of of soundsthat humans can hear is about
200 decibels and that's fromlike absolute dead silence to
your head explodes.
So right, like we've got plentyof usable range here.
So yeah, if your signal goes200 db above zero, yes, the
(13:59):
points it's jumping to are muchfurther apart and also, that's
so much louder than anythingthat might be closer together
that it's going to completelyswamp it.
So you kind of you don't need tocapture stuff that's at 200 and
that's at negative 18 at thesame time.
There just there's no, and sothat's kind of the rationale
(14:19):
behind it and why it can't.
Same thing when you go theopposite direction, right, the
spacing gets closer together.
But if your signal is down atnegative 80, you've got to be
able to distinguish that fromother noise things and usable
stuff.
So by time you magnify thatback up to a usable range.
Now your noise is actually waydown there and your usable stuff
(14:40):
is now at negative 18 orwhatever I see that makes sense,
so it's, it's yeah the, theranges between the points that
you snap to change.
They get bigger and smaller butthe relative signals that you're
trying to capture at thoselevels it kind of captures
what's important.
It has enough range in any onekind of zoom factor to capture
(15:01):
what's important and it can kindof get rid of the rest.
Right, it cannot capture stuffthat's not important yeah, but
that's something that thatpretty much exclusively happens
inside a DAW.
There I am starting to see, um,there are field recorders
specifically that record to32-bit float.
To my knowledge and this is, Ihave to admit, this is something
(15:24):
I need to get smarter on to myknowledge, they actually have
multiple 24-bit converters, ormaybe even lower, that they're
switching between on the fly,dynamically based on the input
level.
So you set your overall gain ormaybe some of them.
You don't even have to set gainand it just based on the input
level, it'll pick the right kindof stage of a 24-bit converter
(15:47):
and it'll compile that into a32-bit float word that you can
then record in 32-bit float.
But it's not really truly a32-bit float converter, so to
speak.
Right, okay?
Marc Matthews (16:01):
Interesting.
Ian Stewart (16:02):
It's what we would call multi-path.
Yeah, yeah, right Okay.
Marc Matthews (16:04):
Interesting.
It's what we would callmulti-path.
Yeah, yeah, yeah, that'sprobably, I think, using the
ruler.
There is a really, really goodway of describing it, I think,
for the audience listening.
If you're watching this, do goand watch this on YouTube as
well, just so you can see thephysical descriptions of it as
well, because it really doeshelp get your head around the
(16:24):
the idea of 32 bit.
So just before I think we'llmove on to the um, the audio
examples shortly, just because Ithink this is going to be
really interesting.
But just to go back to the 32bit flow, the magnification
process, so we're using decimals, can you just?
It's?
(16:45):
I appreciate.
I don't know if there is ahigh-level overview of the sorry
, not the magnification, themultiplication of that or what's
happening there.
Ian Stewart (16:55):
Unfortunately, the easiest and best way to describe
it probably still feels like apain to most people, especially
people in the arts, but it'sbasically scientific notation,
right?
So if you remember scientificnotation of numbers, so you
might say 1.278 times 10 to thesecond and that basically means
(17:21):
you move the decimal place overtwo times 10 to the second is
100, so it's times 100.
10 to the third is 1,000, soit's times 1,000.
So it's basically sort of likestoring audio in scientific
notation and that's how thatscaling is happening.
So it's determining, kind ofyour zoom factor.
That's, admittedly, a littlehand wavy and there are actually
(17:44):
different formats.
Uh, I will.
There's actually there is agood article I found that kind
of explains this.
You have something called themantissa and the oh, I'm
forgetting the other part.
I'll try and send that to youafter mark if you want to
include it on on like show notesor something.
Exponents mantissa exponentsyeah, or possibly um yeah, but
(18:07):
yeah, yeah, um significant andyeah, whatever, yeah but, it's
and there's.
There's an ieee specification.
I think it's, I'm not gonnaguess.
I'll send you the link and andthere's a link.
There's a link that I foundthat describes it pretty well.
It's a little mathy, but itkind of helps you get your hand
around it no, that'd beinteresting.
Marc Matthews (18:25):
Um, send it across an audience listening.
If you want to dive deeper,I'll put a link to it in the
episode description and you can.
You can go and have a listen.
But yeah, and I think we'vebeen, we've been teasing these,
these audio examples.
I think let's dive into thesethen.
So, uh, this is new territoryhere, folks.
We've done audio before, butI've never shared my screen on
(18:45):
the podcast, so we'll describewhat we're viewing as we go
through this.
Obviously, you're going to hearit, but if you want to see it,
do go check out the YouTubevideo.
So I'm going to share my screen.
I'm moving my head to the lefthere because I need to look away
from my camera.
Ian Stewart (19:01):
And Mark, while you do that, I'll just a little
kind of one other thing.
I think if there is a mytharound dither, it's that it
masks the distortion or itcovers up the distortion.
That otherwise happens.
That is not true.
It actually replaces it, andwe're going to see that.
(19:23):
So, mark, actually I don't knowif there's a way, if you can
put that.
Well, let's see, you're onnumber four.
Marc Matthews (19:31):
I wanted to change.
Do you want me to change?
Ian Stewart (19:37):
Yeah, at the very top on your left bar.
There you've got number five.
Is it possible to just put thatat the very bottom?
Marc Matthews (19:40):
Yeah, yeah there we go.
Does it move to the bottom, oreven maybe move that one?
Ian Stewart (19:46):
there we go, there we go all right, so what we can
start with.
If you want to look at numberone here, mark faded low level
tone faded low, here we go sothis is a sine tone.
It's wobbling, uh plus and minusa little bit around 1k and it
is very quiet.
It.
(20:06):
It is very, very quiet.
That marker.
There's a marker for if you'rewatching on YouTube, there's a
marker there that's negative96.3 dBFS.
That is the theoretical noisefloor of 16-bit.
And so I put that marker in todemonstrate where that wobbling
sine tone crosses negative 96.
And so this is stored in 32-bitfloat.
(20:28):
In fact, actually this is ascreenshot of RX If you look at
the very bottom of the bar, kindof slightly to the left-hand
side, you can even see it says32-bit float, 44,100 hertz, 44.1
kilohertz, right, so this isour sine tone that we're
starting with.
And then, Mark, let's look atnumber two.
(20:49):
This is what happens if we dowhat's called truncation, so we
just chop off the bottom bitsand go down to 16 bit, and so,
for those of you looking at this, you can see there's all sorts
of tones above that kind of looklike harmonics a little bit,
(21:10):
and sometimes they can have aharmonic relationship.
They don't always have to.
But the other thing that yousee is all this other junk in
the background.
That is sometimes when the toneis going up, the junk in the
background is going up, andother times it's going down in
the opposite direction.
It's, it is, it is very random.
And so what aliasing is?
(21:32):
Uh, not aliasing, sorry, whatdither is doing.
So this is, these kinds ofartifacts that we're seeing are
what we would call truncation,distortion.
And, mark, let's actually, ifwe could, could play.
There's a file, I think it's.
It should say fade plus 60 db.
No, dither yeah, I've got thatum, so here we go, okay that's
(22:05):
the audio of the image thatwe're looking at and that should
just be a very pure tone and itshould fade out.
And it's kind of neither ofthose things, right.
There's a lot of backgroundjunk that you can hear.
It's not always just it's doingrandom stuff in the background
and it also the tone kind ofstays at the same level and then
stops right.
(22:25):
So it's not fading and it's notjust a pure tone in the
background and it also the tonekind of stays at the same level
and then stops right.
So it's not fading and it's notjust a pure tone.
So I don't want to go too deepwith this, but let's just talk
very quickly about what'sactually happening and why we
add dither and what it's helpingdo.
So when this tone gets reallyquiet, we have what's called an
(22:47):
LSB least significant bit andthat's basically, if we're on a
ruler, that's in millimeters,it's the distance between zero
and one millimeter.
It's that last tiny measurabledistance before we get to zero.
And when the signal is veryquiet, what happens is it
(23:07):
basically toggles on and offbetween those two right.
Sometimes it toggles to zero,it gets rounded down to zero and
other times it gets rounded upto one millimeter and that's not
actually representative of whatthe audio is doing.
And at some point the audiogets quiet enough that it's less
(23:28):
than half a millimeter awayfrom zero and it just all gets
quantized to zero and thattechnically I mean it's silence.
That's what happens at the endof this file, when it just stops
and on the screen you can seeit just goes to black.
But that's also technicallyinfinite distortion at that
point because there's stillsignal there.
It's just been removed anddeleted entirely.
(23:51):
And so what dither does?
Dither is this very quiet noisethat we add, and noise.
One of the things about noiseis that it's random, and so by
adding this noise sometimes itbumps the signal up a little bit
, so it maybe quantizes to onemillimeter, maybe sometimes it
quantizes to two, other times itbumps it down, so maybe it
(24:13):
quantizes to a negative one,right If we extend the ruler to
go into the negative direction.
And so by adding the noise itadds this random element to it
that actually can now retain.
So what it does is technicallyit decorrelates the quantization
distortion from the signal andnow it literally replaces that
(24:37):
distortion with noise.
So now you have noise and yourlow level signal can ride on top
of that noise, completely,perfectly retained.
So, mark, let's look at the nextimage there, image three, and
this is an image of addingdither, and now you can see that
we just have that pure tone,that it continues actually below
(25:01):
that negative 96.3 mark, right,that tone keeps going until it
really just fades down into thenoise.
So we see this kind of noisebehind it and then what you're
seeing up at the top, the kindof brighter orange area.
This is because we're doing Imentioned that thing noise
shaping earlier.
Basically, it's like applyingan EQ to the noise, and so what
(25:22):
we can do is, because our earsare much more sensitive between
about 1k and 10k or 8k, we canEQ the noise so that we make the
noise quieter where our earsare more sensitive and louder
above, like 16k, where our earsare less sensitive, right, and
that's noise shaping.
We don't even really hear thenoise as much as we would
(25:44):
otherwise.
So, mark, now let's play that.
Play the next audio file,that's the fade plus 60 to be
with dither.
So now you'll hear a littlenoise in the background, but
just the pure tone, and you'llactually hear it fade out.
Here we go All right.
(26:09):
So it just fades into the noisethere until it kind of the
noise does mask it.
But what the noise is maskingis the original signal, not
distortion, right, thedistortion is actually gone and
replaced with noise.
(26:31):
Now, the other thing that wehave to say, that you can kind
of see on the screen, these areboosted by 60 dB.
Right, these are very quietthings that we're normally
talking about, and so there's alittle it's a little maybe
deceitful of us, but I thinkit's useful to hear it in this
very clear way of the tonaldifference and the fact that
really the dither is removingand replacing that distortion
with just constant state noise.
And if you drop this by 60 db,you would just kind of hear that
(26:54):
tone fade out.
But if you drop the, the tonalone by six, or the, the
truncated one, by 60 db, youmight then start hearing some of
those alien, those truncation,distortion artifacts yeah, but
also it would cut it off as well.
Marc Matthews (27:08):
And it would just cut off, you wouldn't get the
fade out.
Yeah, yeah.
Ian Stewart (27:10):
Yeah, so we have two more images and audio
examples.
This is just a snare drum andI've gone to 8-bits again.
That's a very low bit depth,but to make it really clear.
So here's a snare drum and,mark, I think I may have even
given you just the 32-bit floatversion.
(27:30):
How many snare drum files arethere?
Marc Matthews (27:34):
I've got three.
I've got two 8-bits and a24-bit.
Ian Stewart (27:39):
So let's ignore the 24-bit, so let's play the 8-bit
, one that's truncated.
Marc Matthews (27:45):
So no dither, here we go.
Ian Stewart (27:47):
Yeah, oh, wow, yeah , you get that reverb sound you
can just hear it gets grainy andkind of craps out and it's very
noisy.
That reverb is kind of veryeven when it's not kind of
crapping out and gettingstuttery.
(28:07):
It's just a noisy reverb.
And now let's play that, thatdithered one, the 8-bit dithered
one, and if you want us flip tothe next image, you can't yeah,
let's do that, let's do that ohwow, yeah, look at that.
Marc Matthews (28:19):
yeah, right, here we go.
Did I Dither?
Yes, you can hear thedifference, can't you?
Ian Stewart (28:30):
You can add, like the tone of the reverb changes,
it's a little clearer.
The full reverb tail is there,and even right now.
Marc Matthews (28:42):
I still hear the dither noise but it's actually
for eight bits.
Ian Stewart (28:45):
That's pretty quiet and that's because we're using
this really extreme noiseshaping where it's really
emphasizing the high frequencies.
But still that's eight bits butit serves to demonstrate that
dither actually really retainskind of important low-level
information.
So that's we.
(29:08):
You know we touched a littlebit on its random noise, that's
randomizing the signal andmaking it not repeatedly
quantized to these values thatcause distortion.
But really, at the end of theday, it's low level noise that
you add that removes distortionand allows low level signals to
actually persist below you know,places where they would
(29:29):
otherwise get cut off.
And and the other thing, thatpart of why I send the snare
drum example.
Sometimes you may hear peopletalk about things self dithering
.
Right, oh, I don't need toapply and dither because it
already has noise in it andthat's not really accurate.
Marc Matthews (29:46):
And if we think about this, well, it's not a
real thing.
Ian Stewart (29:50):
That's the thing.
People talk about it as if itis.
I mean, I guess maybe there area few circumstances where it
could be.
But part of why I send thesnare drum is like it seems like
the snare hit and then thereverb is noise that you're
adding Like the reverb is thisnoisy signal because the snare
is a noisy signal to?
Is noise that you're addingLike the reverb is this noisy
signal because the snare is anoisy signal to start with.
So it kind of seems likethere's noise, so shouldn't it
(30:11):
just dither?
And the key is that the noisehas to be completely
uncorrelated, unrelated from thedigital audio that's coming in,
right?
So just because a plugin hassome noise or whatever doesn't
mean that you can get away withnot dithering um, once, once
(30:31):
that's part of the signal and itdrops below these low levels.
You need to dither to to kindof recover and or retain that
low level information yeah, Ithink these, these examples are
brilliant.
Marc Matthews (30:45):
Uh, audience listening.
Do please, if you, if you don'tusually watch the podcast, do
go find it on youtube, thisepisode 204, just so you can
listen and then see, becauseit's really interesting when you
see it in rx and you canactually see the, the audio file
itself and and what ian isdescribing and it really does
help cement what's happeninghere.
(31:05):
Quick question about noiseshaping.
So you mentioned there oursensitivity and a particular
frequency, sort of bandwidth,and then we can use noise
shaping within our DAW to do it.
I was going to say de-emphasize, if that's a word.
Yeah, kind of I suppose it is inthat particular range where
(31:27):
we're most sensitive, and is itemphasizing it further in the
higher frequencies as a result?
Ian Stewart (31:34):
Yeah, mostly high frequencies and a little bit
lower frequencies, yeah Right,so it's kind of.
Actually you could see what thenoise-shaping curve was.
Mark, you don't need to pullthe pictures back up, but if
anyone wants to rewind and lookat those, the dither module in
RX was kind of in the lowerright and there's a yellow line
that shows basically what the EQshape of the noise shaping is.
(31:56):
So yeah, it's kind of doing abroad dip from about 1K usually
up to about 12K or so, andthere's a little bit of a low
shelf just very little, and thena really big high shelf that
boosts the noise up, kind of,you know, as it approaches 20k
and you can do different amountsand and the the interesting
thing with dither is, uh, thethe total power of right we can.
(32:23):
We can think about peakamplitude or the power of a
signal and really.
So the average power of adither signal, a noise that is
valid dither, has to be a prettyprecise level to work.
I mean, it can be louder thanthat, but then it's just more
noise than you need.
(32:43):
But if the noise gets too quietthen it stops working and you
run back into truncation,distortion.
So the power of that signal hasto be pretty much very precise.
So when you apply noise shaping, the absolute power of that
noise is the same.
It's just shifting where thatenergy is in the frequency band.
Marc Matthews (33:04):
Right, I get you, I get you, I get you Right.
Ian Stewart (33:07):
Interesting stuff.
So you can make the averagepower in that really sensitive
range be lower by shifting a lotof that up high or a little bit
down low so it sounds quieter,but it's actually technically
the same average level.
Marc Matthews (33:22):
The same.
Yeah, that's really interesting, really interesting, because
you see, in modern DAWs thereare various different dithering
options.
Options, yeah, available to us.
Ian Stewart (33:32):
Yeah.
Marc Matthews (33:33):
So I think we've gone through and we've gone
through the examples of ditherand what it is why we need to be
aware of it.
So, when it comes to actuallyusing dither, what is your top
advice?
So you're in your door, you'reabout to export a mix and, yeah,
(33:53):
what is your best advice when?
It comes to exporting a mixjust in general.
So the.
Ian Stewart (33:58):
The first rule of dither is anytime you're
reducing the bit depth uh,sometimes you might hear people
refer to that as word length.
Bit depth and word length arereally basically the same thing.
Anytime you're reducing that,you need to add dither.
That that is like the guidingrule, right?
I mentioned earlier.
Actually, when you're recording, you're theoretically reducing
(34:23):
from infinite bit depth of ananalog signal to 24-bit or
whatever.
So your interface is addingdither at that point to even
record properly.
But anytime you're reducing thebit depth of a signal, you
should add dither.
So whether you're going from 32float down to 24 bit or 24 bit
down to 16 bit, those are thetimes that you should add dither
(34:44):
.
If you go from 32 floatstraight to 16, you should add
dither.
If you go from 16 up to 24, youcan do that, but you're not
adding any real signal.
There's no real benefit indoing that.
So you don't need to then adddither.
You're just sticking eightzeros on the end of all your
samples, basically the one kindof I'm not going to say caveat.
(35:13):
But edge case here is if youtake something that's 16-bit got
.
But edge case here is if youtake something that's 16-bit and
you bring it into your DAW andyou do literally anything to it
0.1 dB of gain, a plugin,literally anything other than
just play it straight back.
Your DAW will then convert itto 32-bit float.
(35:33):
And so if you do any furtherprocessing, even if you've
dithered to 16-bit,unfortunately that then means
that you should dither again,and that's a real good reason
not to go down to 16-bit untilthe very end of the chain, when
you're done mastering, and eventhen only if you're really going
to CD, because these days wetalked about this last episode,
(35:53):
right, about this last episode,right you can distribute 24 bit.
So, um, very practicallyspeaking, what I tell or ask or
suggest, not tell.
I don't tell anyone to doanything.
Do what you want.
You're an intelligent personwho can make decisions.
You're a free person, I hope,um, but what I suggest is that,
(36:17):
actually, if you're sending memixes, you might as well just
send them that 32-bit float.
That's fine, because I'mprobably going to adjust the
gain anyway.
I've got to go out to my analog.
Well, I don't have to, but mostof the time I end up going out
to my analog chain.
So, for a mix, you can export amix at 32-bit float.
Avoid dither.
I'll add it when I'm done.
Life's good, uh.
But if you want to go to 24, 24bit, um, fixed point, uh,
(36:42):
whether it's a size thing, oryou just want to go play it out
and listen to it, uh, you should.
You should then dither too At24 bit.
Theoretically, the noise levelis low enough that noise shaping
doesn't really make an audibledifference.
You can't really hear it, soyou can just use.
A lot of times, you'll see.
(37:02):
The option will be TPDF.
If you've ever wondered whatthat stands for, here you go
Triangular Probability DensityFunction.
We're not going to go anydeeper than that, but basically
it just means flat white noise.
Yeah, yeah, so at 24-bit,that's okay.
Actually, though reallyinteresting and relevant to this
(37:24):
, bobcats legendary masteringengineer, alexi Lukin, lead DSP
engineer from iZotope, and I dida series of tests a while ago
and I actually need to reachback out to them and kind of
keep digging into this uh wherewe tested some different dither
shapes and how different thingsrespond to them, and alexi, I
(37:46):
think, found that noise shapedither, even at 24 bit, provides
kind of more resilience, I'mgoing to say, to digital volume
controls.
So we're getting into slightlyweird heady territory here, but
just briefly quickly.
I'll try and keep it high level.
When we turn the volume up anddown on devices that we listen
(38:09):
to, sometimes it may be analogand it's literally just an
analog potentiometer that'spadding that signal down from
the maximum level that comes outof the amplifier.
That's great, that's kind ofthe ideal, because you come out
of the converter at 24-bit withdither and then you just trim
that analog signal.
But a lot of devices, phones,bluetooth we're back to the
(38:31):
Bluetooth headphones from lastweek a bunch of devices have
digital volume controls andthere they're actually doing
they convert to 32-bit floatagain.
So like, why the hell don't wejust distribute that 32-bit
float?
Well, because it doesn't alwayshappen.
So we still have to dither anddistribute at 24-bit.
But sometimes digital volumecontrols are a floating point
(38:52):
and then they have to reconvertdown to 24-bit to get it to the
converter.
And what we found in our testsis that actually those signals
survive better and are moreresilient to truncation
distortion when they go throughdigital volume controls.
If you noise shape that 24-bitdither, so it doesn't really
make an audible difference untilyou start adjusting the level
(39:14):
and then some weird stuff canhappen.
So if you're just sending itoff for mastering tpdf.
Just the regular flat, un-noiseshaped dither at 24-bit is
totally fine for the masteringcrew out there.
If you're thinking aboutsending out 24-bit files, you
may want to start noise shaping,just because there can be some
benefits yeah, well, youmentioned that about the um,
(39:38):
sorry, but no, no, no, it'sfantastic about the 32-bit.
Marc Matthews (39:42):
When you're again , it's not ever some.
When I, when I bring a, a stem,a sample, whatever it may be,
into a daw project that I'mworking on, never crosses my
mind to think, okay, that'schanging the sample, um, sorry,
not the sample, right, it'schanging the um, the bit depth
of that file.
You just, yeah, it just nevercrosses my mind, it's never
something I think about.
(40:02):
And then, like you said there,if you're working at 32 bit,
it's then going to change it to32 bit, or, I suppose, if you
work with 24, but it's going todo the same again if it and and
actually I mean here's aninteresting thing Different DAWs
behave a little bit differently, but for example in Ableton.
Ian Stewart (40:23):
Ableton pretty much does everything at 32-bit flow
and it's really only once youexport that those well actually
I take that back.
Your project settings inAbleton affect your recordings.
So if you record stuff in froman interface, that's where that
bit depth comes in It'll usethat.
Once you're inside, if you doany flatten and freeze or things
(40:45):
like that, that'll all stay at32 float and it's only once you
then export and select your bitdepth, that that factors in.
Again Pro Tools is a littledifferent.
That factors in.
Again.
Pro Tools is a little different.
Pro Tools if you set Pro Toolsto 24 bit and you print
internally or you do differentbounces, it'll convert down from
32 float and there aredifferent rules about when it
(41:11):
uses dither and doesn't, and Ican never keep it all in my head
and so practically I have tosay that I don't use pro tools a
lot, but actually I use it withmy students at Berkeley and
when we do I just tell them lookfirst thing when you go in, if
you're doing a mastering session, because we're going to be
printing, we're going to besending stuff between auxes and
doing different versions ofstuff.
Just set it to 32 bit float.
That way everything internal isgoing to stay at floating point
(41:32):
until you explicitly go out andput a dither on your output and
say I'm going down to 24-bitand these are my settings, and
now you know, export at that bitdepth, but yeah, it's a little
bit of.
The other thing is like you know, I don't know who I first heard
say this, I don't know if itwas Ian Shepard or Bob Katz or
(41:55):
someone entirely different, butno one ever ruined a good
recording by not dithering Right.
That's the other thing to sayon the list of really important
things to do when you're tryingto make your music sound great.
Dither is probably at the verybottom and also it is the
technically correct way to dealwith reducing bit depth and
(42:17):
audio.
So in my view, you should justdo it.
It's the correct way to dealwith reducing bit depth and
audio.
So in my view, you should justdo it.
It's the right way to do it.
So just do it, don't thinkabout it.
It just becomes part of yourstandard process and you know
that you're retaining theselow-level signals and what we
mean by that.
When I say low-level signals,it's stuff like the ambience in
the end of reverbs.
Yeah, which we heard in thesample, which like the ambience
(42:37):
in the end of reverbs, yeah,which we heard in the sample,
which we heard in that sampleand granted, that was at 8 bits,
but even at 16, like, do somelistening tests right, go
between there are.
It's a little hard to do inreal time, but there's a plug-in
by Goodhertz called Good Ditherthat's I think it's like 20 or
30 bucks.
It's very affordable and it'sactually a really excellent
dither algorithm in terms ofbalancing noise shaping and
(42:59):
overall power and all this stuff.
And you can go from no dither,you can set the bit depth and
then go from no dither to noiseshape, to flat, to whatever, and
so if you actually want toaudition that, it's a good way
to do it.
And so you can flip something.
You can set it to 16 bit andthen turn the actual dither on
between the different levels andlisten to it.
(43:19):
And to me.
You know, in a full-on pop orrock mix that's watching my
language here just leave it atfull-on right All the way
through.
Okay, maybe it's not going tomake a huge difference, but as
soon as you start getting alittle dynamic variation or
you've got some room for reverbsto to kind of ring out, or
(43:40):
you've got some softer sections,it can make like it's just.
It's a depth thing.
It just changes a little bitand it just feels a little more
spacious and like there's alittle more room.
And so it does.
It makes an audible difference.
But again, it's not.
It's I wouldn't say it's at thetop of the list or really even
near it, but it's the rightthing to do.
(44:00):
So you know, not worth debatingreplace distortion with noise,
keep a little more depth in yourstuff and just do it at the end
.
Marc Matthews (44:08):
I think that's a perfect way to wrap it up.
I think it, I think it makesperfect sense and I think this
is again going back to likeepisode 203,.
This has been fantastic, andeven more so because we've got
the examples in here.
So again, audience, please dogo check out the video so you
can see the accompanying visualsfor this as well, because it
really does help cement what Ianhas just described there.
(44:32):
So thank you again for joiningme on this.
I think this is the fourthepisode, fourth time you've been
on on the podcast and firstpodcast I've ever been on.
Ian Stewart (44:40):
Four times so amazing records on both sides.
Marc Matthews (44:44):
Yeah, yeah, most definitely mark.
Ian Stewart (44:46):
I really appreciate it.
I I love chatting about thisstuff.
Marc Matthews (44:49):
As you can probably tell, I'm an I'm an
uber nerd when it comes to it,so thank you for giving me a
platform to nerd out with peoplethat hopefully appreciate it,
and it's really it's fantasticbecause it's kind of whenever I
do get those questions, I'mthinking you know what I I'm not
even going to try to attempt todescribe it in on the podcast I
can reach out to you that Iknow a guy who can do it.
(45:11):
Yeah, uh, he can do it verywell, uh, and I could reach out
to yourself and get you on theshow.
So I know we've already spokenlast in episode 203, about
potential future episodes, sowe'll just do keep your ears and
eyes out for that one there aswell.
Ian, again, for those who maybe listening to the podcast for
the first time, where canaudience find you online?
(45:34):
Learn a bit more about whatyou're doing?
Ian Stewart (45:37):
Yeah, absolutely Best place, I always say, is my
website flotownmasteringcomF-L-O-T-O-W-N.
Masteringcom.
There's links, there's a page Ithink it says words is the
heading, and there's spoken andwritten.
So there's links to my blogentries and all the podcasts
that I've been a guest on, blogentries and all the podcasts
(45:59):
that I've been a guest on.
Um, and you know, if you wantto work together on some
mastering stuff, that's reallymy favorite thing to do.
Uh, so there's an intake form.
You can.
You can get in touch and sendme music if you want to work on
something.
Uh, ian Stewart music onInstagram.
Uh, you know, it's mostlypictures of me hiking with my
dogs and stuff like that, butyou know I'll throw a release
that an artist, the client, hasjust done.
If they tag me and stuff, I'llthrow that in my story so you
(46:20):
can see some of the stuff I'mworking on and stay in touch.
Marc Matthews (46:22):
So yeah, fantastic stuff.
Links will be in the episodedescription, audio audience
listening.
So please do go and check thatout.
And, as I mentioned at thebeginning, don't forget, if you
want to make those improvementsto your music playlist worthy
music as well do sign up for myweekly newsletter, where you get
tips and tricks direct to yourinbox.
(46:43):
For example, uh ian's uh wavelabs series was featured
recently in that newsletter.
So please do get signed up forthat.
Entirely free, no fluff, nospam.
Just once a week you'll get anemail from me and uh ian.
It's been an absolute pleasure.
We will do it again.
I'm sure we will.
As I mentioned, we've got othertopics to go through.
So again, it's been an absolutepleasure and folks, until next
(47:06):
time, stay inspired, keepcreating, don't be afraid to
experiment inside the mix.
Ever hear the term dither in audio production
and wonder do I actually needthis, or is it just another myth
?
I think myth is probably thewrong word there, because it's
not a myth, it does exist.
What if I told you that addingnoise to your final bounce could
actually improve the way yourmusic sounds?
Well, that's exactly what we'rediving into today.
Ian Stewart (00:22):
You're listening to the Inside the Mix podcast with
your host, Mark.
Marc Matthews (00:26):
Matthews, welcome to Inside the Mix, your go-to
podcast for music creation andproduction.
Whether you're crafting yourfirst track or refining your
mixing skills, join me each weekfor expert interviews,
practical tutorials and insightsto help you level up your music
and smash it in the musicindustry.
Let's dive in.
(00:46):
Hey folks, welcome to Insidethe Mix.
This is part two of a specialseries with mastering engineer,
educator and plugin developer,ian Stewart.
In part one we explored samplerates and busted some big myths
around digital audio.
If you haven't heard that oneyet, be sure to check out
episode 203.
And today we're zooming in on atopic that's even more, I say,
(01:10):
misunderstood.
I guess it could be wellmisunderstood, and that is
dither.
So a bit about ian, if this isyour first foray into inside the
mix.
He's worked with um artistssuch as krs1 and countless indie
artists.
He co-developed a baseline proand teaches mastering at
Berkeley as well.
So you know you're learningfrom someone who lives and
(01:30):
breathes mastering.
He lives and breathes thisstuff.
Ian, welcome back.
I'm going to say how are you?
But we're recording theseepisodes and series at the same
time, but I'll ask it anyway.
How are you?
I'm great.
Ian Stewart (01:43):
As you can see, I changed between it.
No, um, yeah, no, it's, it's.
Marc Matthews (01:53):
I'm actually getting a little bit of a breeze
through the window here, whichis really, really lovely on this
balmy new england day.
Yeah, I am jealous because inin my studio here I don't have
uh windows.
Ian Stewart (02:00):
It is essentially no windows in a mastering room
are weird, but that's whathappens when you turn a guest
room in your house into yourmastering room yes, yeah, yeah,
I.
Marc Matthews (02:09):
When it gets to the summer, I'd be thankful.
Instead, I've got a a a heaterthat has a cooling feature on it
, so I'm having to use that.
It's pretty grim, it's prettygrim and it gets hot in here
towards the end of the day.
Ian Stewart (02:19):
Yeah, no, once I have the, the tv and all the
analog gear switched on, it'sjust yeah, it gets absolutely
brutal in here.
Marc Matthews (02:26):
I bet man.
So this episode is going to beyour go-to guide for
understanding what dither is,how it works and why it matters.
So we're going to be talkingabout when to apply it, what
happens if you skip it andwhether it makes an audible
difference in real-worldscenarios.
So if you've ever explored yourmix wondering should I hit the
dither button or not, this oneis for you.
(02:47):
But before we dive into that,if you want weekly tools and
tips to your inbox to make yourmusic playlist worthy, click the
link in the episode descriptionand get my free weekly tips to
elevate your music production.
I say my weekly tips not justmine, but I also get guest
contributors as well.
No fluff, no spams.
Just once a week you'll getthose to your inbox.
So in when we're talking aboutdither, I think, as we did with
(03:10):
sample right, it's probably agood place to start.
What is dither and why is itused?
Why do you?
Ian Stewart (03:15):
yeah, so most simply I mean you kind of said
it in your in your intro theredither is noise.
It is just broadband noise,usually kind of white noise,
although we can shape it.
We can use something callednoise shaping.
We'll get more into that, I'msure.
But it's very low level, veryquiet, and it is a signal that
(03:40):
you add in when you are, signalthat you add in when you are
technically, when you arequantizing audio.
So I'm going to refer back tolast week slash 40 minutes ago.
Yeah, yeah, time is an illusion, lunchtime doubly so.
In the words of douglas adams,I think.
Um, so we talked about takingsamples 44,100 times or 48,000
(04:06):
times a second right.
So we can do that.
We can slice up the, the audioas it's coming in from our
microphone through our mic preinto our converter.
We can slice it up and then wehave to measure the voltage, and
that part is all pretty muchcompletely.
There's no loss there.
Right, the time slices are whatthey are.
(04:29):
You don't have to compromise onthem.
You can kind of make themwhatever rate you want.
There's just a clock that runsthem very precisely.
But then when you measure thevoltage, you have to turn that
into a digital value and youdon't have infinite numerical
values that you can store indigital.
And so the more kind ofresolution or the bigger a
(04:50):
number you want to be able tostore in digital, the more bits
you need to use.
So one bit can represent twovalues, zero, or one.
Two bits can represent fourvalues, zero, one, two, 3, 4.
It's kind of an exponential-ishthing.
We won't go too deep into that,right, but more bits equals
(05:10):
more values.
So at 16-bit, which was the kindof standard for CD, I say was,
it still is, cds are and everwill be 44.1 kilohertz, 16-bit.
But we don't use CDs as muchthese days.
I'm sure some people still do,but they're they're hardly as
common.
Um, 16 bit gives us enoughvalues that the kind of residual
(05:37):
noise that's left over and weagain we can unpack more of this
.
I'm trying to keep it highlevel at first and we can go
deeper, of course.
Yeah, yeah, um, the residualnoise from kind of having to
round your signal up or down tothe closest level is pretty low.
It's down around 96 DB belowpeak level, right, your clipping
(05:59):
point.
And that's, if you've everheard that the dynamic range of
16-bit audio or CD audio is 96dB of dynamic range.
That's where that comes from.
But basically, right, when youmeasure that voltage, it may
land right at a sample value,right at one of those digital
numbers that you can store, butit could be anywhere in between.
(06:19):
And if it's in between, youhave to either round it up or
down.
You round it to the closestvalue.
And when you do that you haveto either round it up or down
you round it to the closestvalue.
And when you do that, you addthere's a little distortion,
there's a little leveldistortion.
That's being added because it'snot exactly right.
Right, the level that you'restoring it at isn't exactly
right, and a lot of people thinkof this as like graininess or
(06:47):
something like that.
Really, what it does is itdetermines the level of noise,
how quiet this like white noisesignal is below the peak level,
and part of that just like lastweek we talked about aliasing
filters right, this low passfilter that prevents high
frequencies from getting in,being built into your converter.
It's not something you have todo, it's just built in.
One of the other things thathappens when you record a signal
.
When I plug my microphone intothe preamp and it goes into the
(07:10):
converter.
Dither is being added to that.
Now I'm recording at 24 bit sothat noise level is even lower.
It's down around negative 144dB below clipping.
But actually dither is addedwhen you record an analog signal
and that is to prevent to makethe distortion that happens be
(07:31):
noise.
It turns this distortion intonoise, it doesn't mask it.
We have some great examplesthat we'll look at and listen to
that really demonstrate this.
But we can do the same thingwhen anytime we reduce bit depth
if we're going from 32-bitfloat down to 24 or 24 down to
16, we can add dither to preventthat reduction in number of
(07:56):
steps that we can quantize tofrom being yucky distortion into
just noise.
That's kind of like tapis andso really it's noise that we add
to the signal.
That prevents another type ofdistortion from happening.
It replaces what we would calltruncation distortion with just
noise and when we have steadystate noise, especially at very
(08:19):
low levels, you kind of justdon't even notice it, whereas
truncation the distortion thatyou get from truncation can be a
little more noticeable and isnot harmonically related to the
music, much like aliasing.
There's no harmonicrelationship there, so it can
just make things sound a littleweird and grainy and gross.
Marc Matthews (08:38):
We've got some examples that we're going to go
through in this episode, whichI'm looking forward to getting
to.
One quick question 32-bit floatyeah, so DAWs now support
32-bit.
So when we're in our DAW, we'reworking away and you see a lot
of discussions again I say thisa lot on the podcast with
(08:59):
regards to DAWs and 32-bit andnot being able to clip and et
cetera when we're using them.
But could you just is there ahigh-level description you can
give of 32-bit and how it'sdifferent to sort of 24 and 16,
if that makes sense?
Ian Stewart (09:17):
Yeah, yeah.
So 16-bit.
24-bit gives you a discretenumber of values that you can.
The word we use is quantize,right, Not average around your
audio levels too, right, youhave to snap them to one of
these levels.
(09:38):
So in the example of if we havetwo bits that can store four
values 0, 1, 2, 3, 4, sorry,zero, one, two, three, really
zero through three right, thoseare four levels.
If, if a level comes in andit's at one half, we have to
make a decision whether to roundit down to one or up to two,
right, so we've got to snap itto one of these levels.
So we call that quantizing, andit's sort of like quantizing
(10:01):
MIDI, right, you're snapping itto the nearest time value.
Same idea, that's why we callit quantizing.
You're snapping it to the kindof the nearest value.
So for 16-bit, we have I thinkit's something like I'm probably
going to get the exact numberwrong, but it's something like
64,000 available values that wecan snap to.
(10:21):
That's a lot.
If that were a stack of paper,uh, it would be very tall, right
?
If each, each piece of paperwere an individual value.
It's like you could do the mathand it's crazy, right?
So already that's giving us alot of values, but it is this
defined bucket of, and and eachvalue is the same, space apart.
(10:43):
Uh, and it's.
It's like, just like if you hada ruler and each, each, each
millimeter marker, or, or 16thinch, or whatever.
You know, we're one of thesevalues, they're all equally
spaced and you have a finitenumber, floating point.
That's so, that's fixed point,that's 16 bit or 24,.
Four bit can have 32-bit fixedpoint, but it's pretty uncommon.
(11:05):
Floating point is a littledifferent.
You have, basically, the way tothink of it is you have a
decimal number and that.
So that would be like 0.
Dot, dot, dot or 1.
Dot, dot, dot or whatever, andyou only have a few digits of
(11:27):
precision there, and then youhave a multiplier, and so what
you're doing is you'remultiplying and scaling this
kind of lower precision thing upand down, and so it's sort of
like you have.
If I go back to the tape measureor ruler analogy, it's like you
(11:48):
have a shorter ruler with lessmarks on it, but you can almost
like magnify it or zoom it upand down.
I'm going to try and stay inmetric.
For most of the world.
It might be, you know, fivecentimeters down here.
(12:09):
But if you scale it way up andyou need it, you know you need
it to store a much larger value.
Now it might be a kilometerlong, but the spaces in between
those marks are also now larger.
So you could have fivecentimeters down here with
centimeter marks and then youcould go up to a kilometer, but
(12:29):
now the marks are 200 metersapart right, so it's does that
make sense?
it does, yeah, describing it itdoes.
Marc Matthews (12:39):
But in my head I'm thinking if, if that so with
regards to audio because youmentioned about the the um, the
tape, measure the ruler down, ifwe if you're watching this on
on youtube, because I'm sort ofpointing down we've got a
narrower range and then furtherup we've got a larger range.
(13:00):
So if we were to go into thatlarger range with audio, how is
that stored?
Does that not mean that we'resnapping to points further apart
?
Ian Stewart (13:12):
We are, and also the signal is a lot louder.
So the relationship right.
So if your signal goes 200 dBabove full scale, which, for
what it's worth, 32-bit float,does have a clipping point and a
noise floor, it's just that theclipping point is something
like 750 dB above zero and thenoise floor is 700, something
(13:34):
else below zero.
So it's about a 1500, 1,500decibel range that you have and
the range of sounds that humanscan hear is about 200 decibels
have and the range of of soundsthat humans can hear is about
200 decibels and that's fromlike absolute dead silence to
your head explodes.
So right, like we've got plentyof usable range here.
So yeah, if your signal goes200 db above zero, yes, the
(13:59):
points it's jumping to are muchfurther apart and also, that's
so much louder than anythingthat might be closer together
that it's going to completelyswamp it.
So you kind of you don't need tocapture stuff that's at 200 and
that's at negative 18 at thesame time.
There just there's no, and sothat's kind of the rationale
(14:19):
behind it and why it can't.
Same thing when you go theopposite direction, right, the
spacing gets closer together.
But if your signal is down atnegative 80, you've got to be
able to distinguish that fromother noise things and usable
stuff.
So by time you magnify thatback up to a usable range.
Now your noise is actually waydown there and your usable stuff
(14:40):
is now at negative 18 orwhatever I see that makes sense,
so it's, it's yeah the, theranges between the points that
you snap to change.
They get bigger and smaller butthe relative signals that you're
trying to capture at thoselevels it kind of captures
what's important.
It has enough range in any onekind of zoom factor to capture
(15:01):
what's important and it can kindof get rid of the rest.
Right, it cannot capture stuffthat's not important yeah, but
that's something that thatpretty much exclusively happens
inside a DAW.
There I am starting to see, um,there are field recorders
specifically that record to32-bit float.
To my knowledge and this is, Ihave to admit, this is something
(15:24):
I need to get smarter on to myknowledge, they actually have
multiple 24-bit converters, ormaybe even lower, that they're
switching between on the fly,dynamically based on the input
level.
So you set your overall gain ormaybe some of them.
You don't even have to set gainand it just based on the input
level, it'll pick the right kindof stage of a 24-bit converter
(15:47):
and it'll compile that into a32-bit float word that you can
then record in 32-bit float.
But it's not really truly a32-bit float converter, so to
speak.
Right, okay?
Marc Matthews (16:01):
Interesting.
Ian Stewart (16:02):
It's what we would call multi-path.
Yeah, yeah, right Okay.
Marc Matthews (16:04):
Interesting.
It's what we would callmulti-path.
Yeah, yeah, yeah, that'sprobably, I think, using the
ruler.
There is a really, really goodway of describing it, I think,
for the audience listening.
If you're watching this, do goand watch this on YouTube as
well, just so you can see thephysical descriptions of it as
well, because it really doeshelp get your head around the
(16:24):
the idea of 32 bit.
So just before I think we'llmove on to the um, the audio
examples shortly, just because Ithink this is going to be
really interesting.
But just to go back to the 32bit flow, the magnification
process, so we're using decimals, can you just?
It's?
(16:45):
I appreciate.
I don't know if there is ahigh-level overview of the sorry
, not the magnification, themultiplication of that or what's
happening there.
Ian Stewart (16:55):
Unfortunately, the easiest and best way to describe
it probably still feels like apain to most people, especially
people in the arts, but it'sbasically scientific notation,
right?
So if you remember scientificnotation of numbers, so you
might say 1.278 times 10 to thesecond and that basically means
(17:21):
you move the decimal place overtwo times 10 to the second is
100, so it's times 100.
10 to the third is 1,000, soit's times 1,000.
So it's basically sort of likestoring audio in scientific
notation and that's how thatscaling is happening.
So it's determining, kind ofyour zoom factor.
That's, admittedly, a littlehand wavy and there are actually
(17:44):
different formats.
Uh, I will.
There's actually there is agood article I found that kind
of explains this.
You have something called themantissa and the oh, I'm
forgetting the other part.
I'll try and send that to youafter mark if you want to
include it on on like show notesor something.
Exponents mantissa exponentsyeah, or possibly um yeah, but
(18:07):
yeah, yeah, um significant andyeah, whatever, yeah but, it's
and there's.
There's an ieee specification.
I think it's, I'm not gonnaguess.
I'll send you the link and andthere's a link.
There's a link that I foundthat describes it pretty well.
It's a little mathy, but itkind of helps you get your hand
around it no, that'd beinteresting.
Marc Matthews (18:25):
Um, send it across an audience listening.
If you want to dive deeper,I'll put a link to it in the
episode description and you can.
You can go and have a listen.
But yeah, and I think we'vebeen, we've been teasing these,
these audio examples.
I think let's dive into thesethen.
So, uh, this is new territoryhere, folks.
We've done audio before, butI've never shared my screen on
(18:45):
the podcast, so we'll describewhat we're viewing as we go
through this.
Obviously, you're going to hearit, but if you want to see it,
do go check out the YouTubevideo.
So I'm going to share my screen.
I'm moving my head to the lefthere because I need to look away
from my camera.
Ian Stewart (19:01):
And Mark, while you do that, I'll just a little
kind of one other thing.
I think if there is a mytharound dither, it's that it
masks the distortion or itcovers up the distortion.
That otherwise happens.
That is not true.
It actually replaces it, andwe're going to see that.
(19:23):
So, mark, actually I don't knowif there's a way, if you can
put that.
Well, let's see, you're onnumber four.
Marc Matthews (19:31):
I wanted to change.
Do you want me to change?
Ian Stewart (19:37):
Yeah, at the very top on your left bar.
There you've got number five.
Is it possible to just put thatat the very bottom?
Marc Matthews (19:40):
Yeah, yeah there we go.
Does it move to the bottom, oreven maybe move that one?
Ian Stewart (19:46):
there we go, there we go all right, so what we can
start with.
If you want to look at numberone here, mark faded low level
tone faded low, here we go sothis is a sine tone.
It's wobbling, uh plus and minusa little bit around 1k and it
is very quiet.
It.
(20:06):
It is very, very quiet.
That marker.
There's a marker for if you'rewatching on YouTube, there's a
marker there that's negative96.3 dBFS.
That is the theoretical noisefloor of 16-bit.
And so I put that marker in todemonstrate where that wobbling
sine tone crosses negative 96.
And so this is stored in 32-bitfloat.
(20:28):
In fact, actually this is ascreenshot of RX If you look at
the very bottom of the bar, kindof slightly to the left-hand
side, you can even see it says32-bit float, 44,100 hertz, 44.1
kilohertz, right, so this isour sine tone that we're
starting with.
And then, Mark, let's look atnumber two.
(20:49):
This is what happens if we dowhat's called truncation, so we
just chop off the bottom bitsand go down to 16 bit, and so,
for those of you looking at this, you can see there's all sorts
of tones above that kind of looklike harmonics a little bit,
(21:10):
and sometimes they can have aharmonic relationship.
They don't always have to.
But the other thing that yousee is all this other junk in
the background.
That is sometimes when the toneis going up, the junk in the
background is going up, andother times it's going down in
the opposite direction.
It's, it is, it is very random.
And so what aliasing is?
(21:32):
Uh, not aliasing, sorry, whatdither is doing.
So this is, these kinds ofartifacts that we're seeing are
what we would call truncation,distortion.
And, mark, let's actually, ifwe could, could play.
There's a file, I think it's.
It should say fade plus 60 db.
No, dither yeah, I've got thatum, so here we go, okay that's
(22:05):
the audio of the image thatwe're looking at and that should
just be a very pure tone and itshould fade out.
And it's kind of neither ofthose things, right.
There's a lot of backgroundjunk that you can hear.
It's not always just it's doingrandom stuff in the background
and it also the tone kind ofstays at the same level and then
stops right.
(22:25):
So it's not fading and it's notjust a pure tone in the
background and it also the tonekind of stays at the same level
and then stops right.
So it's not fading and it's notjust a pure tone.
So I don't want to go too deepwith this, but let's just talk
very quickly about what'sactually happening and why we
add dither and what it's helpingdo.
So when this tone gets reallyquiet, we have what's called an
(22:47):
LSB least significant bit andthat's basically, if we're on a
ruler, that's in millimeters,it's the distance between zero
and one millimeter.
It's that last tiny measurabledistance before we get to zero.
And when the signal is veryquiet, what happens is it
(23:07):
basically toggles on and offbetween those two right.
Sometimes it toggles to zero,it gets rounded down to zero and
other times it gets rounded upto one millimeter and that's not
actually representative of whatthe audio is doing.
And at some point the audiogets quiet enough that it's less
(23:28):
than half a millimeter awayfrom zero and it just all gets
quantized to zero and thattechnically I mean it's silence.
That's what happens at the endof this file, when it just stops
and on the screen you can seeit just goes to black.
But that's also technicallyinfinite distortion at that
point because there's stillsignal there.
It's just been removed anddeleted entirely.
(23:51):
And so what dither does?
Dither is this very quiet noisethat we add, and noise.
One of the things about noiseis that it's random, and so by
adding this noise sometimes itbumps the signal up a little bit
, so it maybe quantizes to onemillimeter, maybe sometimes it
quantizes to two, other times itbumps it down, so maybe it
(24:13):
quantizes to a negative one,right If we extend the ruler to
go into the negative direction.
And so by adding the noise itadds this random element to it
that actually can now retain.
So what it does is technicallyit decorrelates the quantization
distortion from the signal andnow it literally replaces that
(24:37):
distortion with noise.
So now you have noise and yourlow level signal can ride on top
of that noise, completely,perfectly retained.
So, mark, let's look at the nextimage there, image three, and
this is an image of addingdither, and now you can see that
we just have that pure tone,that it continues actually below
(25:01):
that negative 96.3 mark, right,that tone keeps going until it
really just fades down into thenoise.
So we see this kind of noisebehind it and then what you're
seeing up at the top, the kindof brighter orange area.
This is because we're doing Imentioned that thing noise
shaping earlier.
Basically, it's like applyingan EQ to the noise, and so what
(25:22):
we can do is, because our earsare much more sensitive between
about 1k and 10k or 8k, we canEQ the noise so that we make the
noise quieter where our earsare more sensitive and louder
above, like 16k, where our earsare less sensitive, right, and
that's noise shaping.
We don't even really hear thenoise as much as we would
(25:44):
otherwise.
So, mark, now let's play that.
Play the next audio file,that's the fade plus 60 to be
with dither.
So now you'll hear a littlenoise in the background, but
just the pure tone, and you'llactually hear it fade out.
Here we go All right.
(26:09):
So it just fades into the noisethere until it kind of the
noise does mask it.
But what the noise is maskingis the original signal, not
distortion, right, thedistortion is actually gone and
replaced with noise.
(26:31):
Now, the other thing that wehave to say, that you can kind
of see on the screen, these areboosted by 60 dB.
Right, these are very quietthings that we're normally
talking about, and so there's alittle it's a little maybe
deceitful of us, but I thinkit's useful to hear it in this
very clear way of the tonaldifference and the fact that
really the dither is removingand replacing that distortion
with just constant state noise.
And if you drop this by 60 db,you would just kind of hear that
(26:54):
tone fade out.
But if you drop the, the tonalone by six, or the, the
truncated one, by 60 db, youmight then start hearing some of
those alien, those truncation,distortion artifacts yeah, but
also it would cut it off as well.
Marc Matthews (27:08):
And it would just cut off, you wouldn't get the
fade out.
Yeah, yeah.
Ian Stewart (27:10):
Yeah, so we have two more images and audio
examples.
This is just a snare drum andI've gone to 8-bits again.
That's a very low bit depth,but to make it really clear.
So here's a snare drum and,mark, I think I may have even
given you just the 32-bit floatversion.
(27:30):
How many snare drum files arethere?
Marc Matthews (27:34):
I've got three.
I've got two 8-bits and a24-bit.
Ian Stewart (27:39):
So let's ignore the 24-bit, so let's play the 8-bit
, one that's truncated.
Marc Matthews (27:45):
So no dither, here we go.
Ian Stewart (27:47):
Yeah, oh, wow, yeah , you get that reverb sound you
can just hear it gets grainy andkind of craps out and it's very
noisy.
That reverb is kind of veryeven when it's not kind of
crapping out and gettingstuttery.
(28:07):
It's just a noisy reverb.
And now let's play that, thatdithered one, the 8-bit dithered
one, and if you want us flip tothe next image, you can't yeah,
let's do that, let's do that ohwow, yeah, look at that.
Marc Matthews (28:19):
yeah, right, here we go.
Did I Dither?
Yes, you can hear thedifference, can't you?
Ian Stewart (28:30):
You can add, like the tone of the reverb changes,
it's a little clearer.
The full reverb tail is there,and even right now.
Marc Matthews (28:42):
I still hear the dither noise but it's actually
for eight bits.
Ian Stewart (28:45):
That's pretty quiet and that's because we're using
this really extreme noiseshaping where it's really
emphasizing the high frequencies.
But still that's eight bits butit serves to demonstrate that
dither actually really retainskind of important low-level
information.
So that's we.
(29:08):
You know we touched a littlebit on its random noise, that's
randomizing the signal andmaking it not repeatedly
quantized to these values thatcause distortion.
But really, at the end of theday, it's low level noise that
you add that removes distortionand allows low level signals to
actually persist below you know,places where they would
(29:29):
otherwise get cut off.
And and the other thing, thatpart of why I send the snare
drum example.
Sometimes you may hear peopletalk about things self dithering
.
Right, oh, I don't need toapply and dither because it
already has noise in it andthat's not really accurate.
Marc Matthews (29:46):
And if we think about this, well, it's not a
real thing.
Ian Stewart (29:50):
That's the thing.
People talk about it as if itis.
I mean, I guess maybe there area few circumstances where it
could be.
But part of why I send thesnare drum is like it seems like
the snare hit and then thereverb is noise that you're
adding Like the reverb is thisnoisy signal because the snare
is a noisy signal to?
Is noise that you're addingLike the reverb is this noisy
signal because the snare is anoisy signal to start with.
So it kind of seems likethere's noise, so shouldn't it
(30:11):
just dither?
And the key is that the noisehas to be completely
uncorrelated, unrelated from thedigital audio that's coming in,
right?
So just because a plugin hassome noise or whatever doesn't
mean that you can get away withnot dithering um, once, once
(30:31):
that's part of the signal and itdrops below these low levels.
You need to dither to to kindof recover and or retain that
low level information yeah, Ithink these, these examples are
brilliant.
Marc Matthews (30:45):
Uh, audience listening.
Do please, if you, if you don'tusually watch the podcast, do
go find it on youtube, thisepisode 204, just so you can
listen and then see, becauseit's really interesting when you
see it in rx and you canactually see the, the audio file
itself and and what ian isdescribing and it really does
help cement what's happeninghere.
(31:05):
Quick question about noiseshaping.
So you mentioned there oursensitivity and a particular
frequency, sort of bandwidth,and then we can use noise
shaping within our DAW to do it.
I was going to say de-emphasize, if that's a word.
Yeah, kind of I suppose it is inthat particular range where
(31:27):
we're most sensitive, and is itemphasizing it further in the
higher frequencies as a result?
Ian Stewart (31:34):
Yeah, mostly high frequencies and a little bit
lower frequencies, yeah Right,so it's kind of.
Actually you could see what thenoise-shaping curve was.
Mark, you don't need to pullthe pictures back up, but if
anyone wants to rewind and lookat those, the dither module in
RX was kind of in the lowerright and there's a yellow line
that shows basically what the EQshape of the noise shaping is.
(31:56):
So yeah, it's kind of doing abroad dip from about 1K usually
up to about 12K or so, andthere's a little bit of a low
shelf just very little, and thena really big high shelf that
boosts the noise up, kind of,you know, as it approaches 20k
and you can do different amountsand and the the interesting
thing with dither is, uh, thethe total power of right we can.
(32:23):
We can think about peakamplitude or the power of a
signal and really.
So the average power of adither signal, a noise that is
valid dither, has to be a prettyprecise level to work.
I mean, it can be louder thanthat, but then it's just more
noise than you need.
(32:43):
But if the noise gets too quietthen it stops working and you
run back into truncation,distortion.
So the power of that signal hasto be pretty much very precise.
So when you apply noise shaping, the absolute power of that
noise is the same.
It's just shifting where thatenergy is in the frequency band.
Marc Matthews (33:04):
Right, I get you, I get you, I get you Right.
Ian Stewart (33:07):
Interesting stuff.
So you can make the averagepower in that really sensitive
range be lower by shifting a lotof that up high or a little bit
down low so it sounds quieter,but it's actually technically
the same average level.
Marc Matthews (33:22):
The same.
Yeah, that's really interesting, really interesting, because
you see, in modern DAWs thereare various different dithering
options.
Options, yeah, available to us.
Ian Stewart (33:32):
Yeah.
Marc Matthews (33:33):
So I think we've gone through and we've gone
through the examples of ditherand what it is why we need to be
aware of it.
So, when it comes to actuallyusing dither, what is your top
advice?
So you're in your door, you'reabout to export a mix and, yeah,
(33:53):
what is your best advice when?
It comes to exporting a mixjust in general.
So the.
Ian Stewart (33:58):
The first rule of dither is anytime you're
reducing the bit depth uh,sometimes you might hear people
refer to that as word length.
Bit depth and word length arereally basically the same thing.
Anytime you're reducing that,you need to add dither.
That that is like the guidingrule, right?
I mentioned earlier.
Actually, when you're recording, you're theoretically reducing
(34:23):
from infinite bit depth of ananalog signal to 24-bit or
whatever.
So your interface is addingdither at that point to even
record properly.
But anytime you're reducing thebit depth of a signal, you
should add dither.
So whether you're going from 32float down to 24 bit or 24 bit
down to 16 bit, those are thetimes that you should add dither
(34:44):
.
If you go from 32 floatstraight to 16, you should add
dither.
If you go from 16 up to 24, youcan do that, but you're not
adding any real signal.
There's no real benefit indoing that.
So you don't need to then adddither.
You're just sticking eightzeros on the end of all your
samples, basically the one kindof I'm not going to say caveat.
(35:13):
But edge case here is if youtake something that's 16-bit got
.
But edge case here is if youtake something that's 16-bit and
you bring it into your DAW andyou do literally anything to it
0.1 dB of gain, a plugin,literally anything other than
just play it straight back.
Your DAW will then convert itto 32-bit float.
(35:33):
And so if you do any furtherprocessing, even if you've
dithered to 16-bit,unfortunately that then means
that you should dither again,and that's a real good reason
not to go down to 16-bit untilthe very end of the chain, when
you're done mastering, and eventhen only if you're really going
to CD, because these days wetalked about this last episode,
(35:53):
right, about this last episode,right you can distribute 24 bit.
So, um, very practicallyspeaking, what I tell or ask or
suggest, not tell.
I don't tell anyone to doanything.
Do what you want.
You're an intelligent personwho can make decisions.
You're a free person, I hope,um, but what I suggest is that,
(36:17):
actually, if you're sending memixes, you might as well just
send them that 32-bit float.
That's fine, because I'mprobably going to adjust the
gain anyway.
I've got to go out to my analog.
Well, I don't have to, but mostof the time I end up going out
to my analog chain.
So, for a mix, you can export amix at 32-bit float.
Avoid dither.
I'll add it when I'm done.
Life's good, uh.
But if you want to go to 24, 24bit, um, fixed point, uh,
(36:42):
whether it's a size thing, oryou just want to go play it out
and listen to it, uh, you should.
You should then dither too At24 bit.
Theoretically, the noise levelis low enough that noise shaping
doesn't really make an audibledifference.
You can't really hear it, soyou can just use.
A lot of times, you'll see.
(37:02):
The option will be TPDF.
If you've ever wondered whatthat stands for, here you go
Triangular Probability DensityFunction.
We're not going to go anydeeper than that, but basically
it just means flat white noise.
Yeah, yeah, so at 24-bit,that's okay.
Actually, though reallyinteresting and relevant to this
(37:24):
, bobcats legendary masteringengineer, alexi Lukin, lead DSP
engineer from iZotope, and I dida series of tests a while ago
and I actually need to reachback out to them and kind of
keep digging into this uh wherewe tested some different dither
shapes and how different thingsrespond to them, and alexi, I
(37:46):
think, found that noise shapedither, even at 24 bit, provides
kind of more resilience, I'mgoing to say, to digital volume
controls.
So we're getting into slightlyweird heady territory here, but
just briefly quickly.
I'll try and keep it high level.
When we turn the volume up anddown on devices that we listen
(38:09):
to, sometimes it may be analogand it's literally just an
analog potentiometer that'spadding that signal down from
the maximum level that comes outof the amplifier.
That's great, that's kind ofthe ideal, because you come out
of the converter at 24-bit withdither and then you just trim
that analog signal.
But a lot of devices, phones,bluetooth we're back to the
(38:31):
Bluetooth headphones from lastweek a bunch of devices have
digital volume controls andthere they're actually doing
they convert to 32-bit floatagain.
So like, why the hell don't wejust distribute that 32-bit
float?
Well, because it doesn't alwayshappen.
So we still have to dither anddistribute at 24-bit.
But sometimes digital volumecontrols are a floating point
(38:52):
and then they have to reconvertdown to 24-bit to get it to the
converter.
And what we found in our testsis that actually those signals
survive better and are moreresilient to truncation
distortion when they go throughdigital volume controls.
If you noise shape that 24-bitdither, so it doesn't really
make an audible difference untilyou start adjusting the level
(39:14):
and then some weird stuff canhappen.
So if you're just sending itoff for mastering tpdf.
Just the regular flat, un-noiseshaped dither at 24-bit is
totally fine for the masteringcrew out there.
If you're thinking aboutsending out 24-bit files, you
may want to start noise shaping,just because there can be some
benefits yeah, well, youmentioned that about the um,
(39:38):
sorry, but no, no, no, it'sfantastic about the 32-bit.
Marc Matthews (39:42):
When you're again , it's not ever some.
When I, when I bring a, a stem,a sample, whatever it may be,
into a daw project that I'mworking on, never crosses my
mind to think, okay, that'schanging the sample, um, sorry,
not the sample, right, it'schanging the um, the bit depth
of that file.
You just, yeah, it just nevercrosses my mind, it's never
something I think about.
(40:02):
And then, like you said there,if you're working at 32 bit,
it's then going to change it to32 bit, or, I suppose, if you
work with 24, but it's going todo the same again if it and and
actually I mean here's aninteresting thing Different DAWs
behave a little bit differently, but for example in Ableton.
Ian Stewart (40:23):
Ableton pretty much does everything at 32-bit flow
and it's really only once youexport that those well actually
I take that back.
Your project settings inAbleton affect your recordings.
So if you record stuff in froman interface, that's where that
bit depth comes in It'll usethat.
Once you're inside, if you doany flatten and freeze or things
(40:45):
like that, that'll all stay at32 float and it's only once you
then export and select your bitdepth, that that factors in.
Again Pro Tools is a littledifferent.
That factors in.
Again.
Pro Tools is a little different.
Pro Tools if you set Pro Toolsto 24 bit and you print
internally or you do differentbounces, it'll convert down from
32 float and there aredifferent rules about when it
(41:11):
uses dither and doesn't, and Ican never keep it all in my head
and so practically I have tosay that I don't use pro tools a
lot, but actually I use it withmy students at Berkeley and
when we do I just tell them lookfirst thing when you go in, if
you're doing a mastering session, because we're going to be
printing, we're going to besending stuff between auxes and
doing different versions ofstuff.
Just set it to 32 bit float.
That way everything internal isgoing to stay at floating point
(41:32):
until you explicitly go out andput a dither on your output and
say I'm going down to 24-bitand these are my settings, and
now you know, export at that bitdepth, but yeah, it's a little
bit of.
The other thing is like you know, I don't know who I first heard
say this, I don't know if itwas Ian Shepard or Bob Katz or
(41:55):
someone entirely different, butno one ever ruined a good
recording by not dithering Right.
That's the other thing to sayon the list of really important
things to do when you're tryingto make your music sound great.
Dither is probably at the verybottom and also it is the
technically correct way to dealwith reducing bit depth and
(42:17):
audio.
So in my view, you should justdo it.
It's the correct way to dealwith reducing bit depth and
audio.
So in my view, you should justdo it.
It's the right way to do it.
So just do it, don't thinkabout it.
It just becomes part of yourstandard process and you know
that you're retaining theselow-level signals and what we
mean by that.
When I say low-level signals,it's stuff like the ambience in
the end of reverbs.
Yeah, which we heard in thesample, which like the ambience
(42:37):
in the end of reverbs, yeah,which we heard in the sample,
which we heard in that sampleand granted, that was at 8 bits,
but even at 16, like, do somelistening tests right, go
between there are.
It's a little hard to do inreal time, but there's a plug-in
by Goodhertz called Good Ditherthat's I think it's like 20 or
30 bucks.
It's very affordable and it'sactually a really excellent
dither algorithm in terms ofbalancing noise shaping and
(42:59):
overall power and all this stuff.
And you can go from no dither,you can set the bit depth and
then go from no dither to noiseshape, to flat, to whatever, and
so if you actually want toaudition that, it's a good way
to do it.
And so you can flip something.
You can set it to 16 bit andthen turn the actual dither on
between the different levels andlisten to it.
(43:19):
And to me.
You know, in a full-on pop orrock mix that's watching my
language here just leave it atfull-on right All the way
through.
Okay, maybe it's not going tomake a huge difference, but as
soon as you start getting alittle dynamic variation or
you've got some room for reverbsto to kind of ring out, or
(43:40):
you've got some softer sections,it can make like it's just.
It's a depth thing.
It just changes a little bitand it just feels a little more
spacious and like there's alittle more room.
And so it does.
It makes an audible difference.
But again, it's not.
It's I wouldn't say it's at thetop of the list or really even
near it, but it's the rightthing to do.
(44:00):
So you know, not worth debatingreplace distortion with noise,
keep a little more depth in yourstuff and just do it at the end
.
Marc Matthews (44:08):
I think that's a perfect way to wrap it up.
I think it, I think it makesperfect sense and I think this
is again going back to likeepisode 203,.
This has been fantastic, andeven more so because we've got
the examples in here.
So again, audience, please dogo check out the video so you
can see the accompanying visualsfor this as well, because it
really does help cement what Ianhas just described there.
(44:32):
So thank you again for joiningme on this.
I think this is the fourthepisode, fourth time you've been
on on the podcast and firstpodcast I've ever been on.
Ian Stewart (44:40):
Four times so amazing records on both sides.
Marc Matthews (44:44):
Yeah, yeah, most definitely mark.
Ian Stewart (44:46):
I really appreciate it.
I I love chatting about thisstuff.
Marc Matthews (44:49):
As you can probably tell, I'm an I'm an
uber nerd when it comes to it,so thank you for giving me a
platform to nerd out with peoplethat hopefully appreciate it,
and it's really it's fantasticbecause it's kind of whenever I
do get those questions, I'mthinking you know what I I'm not
even going to try to attempt todescribe it in on the podcast I
can reach out to you that Iknow a guy who can do it.
(45:11):
Yeah, uh, he can do it verywell, uh, and I could reach out
to yourself and get you on theshow.
So I know we've already spokenlast in episode 203, about
potential future episodes, sowe'll just do keep your ears and
eyes out for that one there aswell.
Ian, again, for those who maybe listening to the podcast for
the first time, where canaudience find you online?
(45:34):
Learn a bit more about whatyou're doing?
Ian Stewart (45:37):
Yeah, absolutely Best place, I always say, is my
website flotownmasteringcomF-L-O-T-O-W-N.
Masteringcom.
There's links, there's a page Ithink it says words is the
heading, and there's spoken andwritten.
So there's links to my blogentries and all the podcasts
that I've been a guest on, blogentries and all the podcasts
(45:59):
that I've been a guest on.
Um, and you know, if you wantto work together on some
mastering stuff, that's reallymy favorite thing to do.
Uh, so there's an intake form.
You can.
You can get in touch and sendme music if you want to work on
something.
Uh, ian Stewart music onInstagram.
Uh, you know, it's mostlypictures of me hiking with my
dogs and stuff like that, butyou know I'll throw a release
that an artist, the client, hasjust done.
If they tag me and stuff, I'llthrow that in my story so you
(46:20):
can see some of the stuff I'mworking on and stay in touch.
Marc Matthews (46:22):
So yeah, fantastic stuff.
Links will be in the episodedescription, audio audience
listening.
So please do go and check thatout.
And, as I mentioned at thebeginning, don't forget, if you
want to make those improvementsto your music playlist worthy
music as well do sign up for myweekly newsletter, where you get
tips and tricks direct to yourinbox.
(46:43):
For example, uh ian's uh wavelabs series was featured
recently in that newsletter.
So please do get signed up forthat.
Entirely free, no fluff, nospam.
Just once a week you'll get anemail from me and uh ian.
It's been an absolute pleasure.
We will do it again.
I'm sure we will.
As I mentioned, we've got othertopics to go through.
So again, it's been an absolutepleasure and folks, until next
(47:06):
time, stay inspired, keepcreating, don't be afraid to
experiment inside the mix.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Special Summer Offer: Exclusively on Apple Podcasts, try our Dateline Premium subscription completely free for one month! With Dateline Premium, you get every episode ad-free plus exclusive bonus content.