Skip to main content

Hey there.
I have some tracks that I have to render to some new tracks cause I have too many plugins going on and my computer is not powerful enough. Now, I've heard that anytime you render a track that's been processed to a new track, you lose some quality because of the re-quantization or whatever. Apparently, dithering helps this, and you can re-process the track as many times as you like if you dither it at the end of the plugin chain. Here's what I don't get: In the Waves' IDR manual, they say that you should dither the track at the same bit depth that it was recorded, but they say you should dither at 16 bit if you want to re-process it, and only at 24 bit if it's the last step of your processing (the mastering stage). Now the thing is, I record at 24 bit, so my questions are:
1- Is there really a quality loss when you render to a new track?
2- If yes, will it help if I dither my tracks before rendering them to a new track?
3- My tracks are recorde at 24 bits, should I dither at 16 or 24 bit?
Thanx in advance.

Carlito

Topic Tags

Comments

JoeH Fri, 02/23/2007 - 13:20

Carlito, here's the short answer:

You only need dithering when bouncing DOWN to a final mix, and only when reducing bit depth, such as 24 to 16. Otherwise, do not do it.

While working within your system, all in the digital domain, one to one "Bounces" will not be compromised if you stay at the current bit level. (For exemple; 24 bit bounces to new, composite 24 bit tracks should be fine, with no gen-loss.)

Of course, any DSP and other processing or combining, etc. will indeed "Change" your original sound, but bouncing per se (at the same bit depth) won't make a bit of difference - it's all digital. (Always save your work and never erase or delete anything until the project is over, and you're sure you'll never need it again....even so, I like to put EVERYTHING - all the parts - onto a hard drive or separate folder for archival purposes.)

Dithering (not to be confused with Sample Rate Conversion - SRC) is used for taking higher res recordings (24/88, 24/96, etc.) down to useable bit depth for things like consumer CDs and video tracks. The short explanation is: Applying a low-level continuous noise source keeps the processor busy and thus more sampling going on, resulting in greater apparent detail in your 16 bit master. (This is why a well-done CD Master can sound so wonderful. You can't nec take it apart and redo it at the 16/44 level, but if it's the last stop in the process, you can relax knowing it's been handled as well as possible.

It's kind've a very good, hi-res digital still photo of a painting/masterpiece, to stretch an analogy. You can start out at 12 meg (or higher) to take the picture with amazing detail and clarity, but as you reduce the resolution, you still have the masterpiece as it appeared, although it won't hold up in its original form should you decide to blow it up again at some point.

"Carlito's Way" was a good flick....Al Pacino, wasn't it?

anonymous Fri, 02/23/2007 - 13:46

Dithering

Thanx a lot JoeH.
Yeah I remember watching Carlito's Way, but that was a longtime ago. I might rent it soon though since you remind me of it.
You seem to know a lot about the bitrate also. So what do you think about recording at a higher bitrate (like 192 khz). Is it really worth it. I tried once to put my soundcard in 192 khz mode (Mackie Onyx 400F). My DAW wouldn't even play. But that might have been some plugin that didn't support that bitrate or something. But since that quadruples the bitrate from 44 khz, I guess it also quadruples the processing power needed right? Is there a big quality difference between 44 and 192?
Thanx again for the advice.

Carlito

JoeH Fri, 02/23/2007 - 14:26

I think you're still a bit confused about bit depth vs. sample rate. Two very different things.

Bit depth is the length of the digital "word" used to describe the sample itself, the number of levels available at a given time. (16 bit yields 65, 536 possible levels, and 24 bit will give 16,777,216 levels. see * below)

Sample rate is the number of times the system "Looks at" or breaks down the audio in question. (So, a 44k sample takes 44,000 snapshots of the audio per second. 88, 96k and upward are doing a lot more than that...)

In most cases, a file's specs is expressed as something like this: 16/44, 24/44 or 24/48, 24/96, etc. THe first number is the bit depth, and the second is the sample rate.

In my own work and experience, I find the bit depth is more important (generally speaking) than the sample rate. I have evolved into working with 24 bits as a matter of routine, in everything I do. It just makes life easier to start at this; all of my DSP's work better, reverbs are smoother, edits and fades as well. (You can also have floating point 32bit going on, but that's another story, for another thread....don't worry about that for now...)

For sample rate for audio work, I find that 44k (and sometimes 88k) is generally more than adepuate, quite excellent, really, esp at 24bits. I like these rates because there is no gearboxing (Sample Rate Conversion - SRC) going on when mastering for CD. The only thing going on is bit reduction (24 down to 16) which is what we're talking about with dithering. For the arguable amount of difference between 44 and 48k sample rate (which I do NOT hear anyway, at 24bits), I'm cool with it.

For video, it's better to work at 48k, since that's what all video systems want to "See". The better systems can now handle either rate with no noticable difference, but again, depending on the targeted project, I like to decide ahead of time if I'm going to be working at 44 or 48k sample rate, and skip the SRC. (And always at 24 bits.)

Clear as mud? :wink:

*Copied from Tweakeadz.com:

Bit Depth refers to the number of bits you have to capture audio. The easiest way to envision this is as a series of levels, that audio energy can be sliced at any given moment in time. With 16 bit audio, there are 65, 536 possible levels. With every bit of greater resolution, the number of levels double. By the time we get to 24 bit, we actually have 16,777,216 levels. Remember we are talking about a slice of audio frozen in a single moment of time.

Now lets add our friend Time into the picture. That's where we get into the Sample Rate.

The sample rate is the number of times your audio is measured (sampled) per second. So at the red book standard for CDs, the sample rate is 44.1 kHz or 44,100 slices every second. So what is the 96khz sample rate? You guessed it. It's 96,000 slices of audio sampled each second.

So lets put it all together now. This brings us to the Bit Rate, or how much data per second is required to transmit the file, which can then be translated into how big the file is. Your CD is 16bit, 44.1 so that is 44,100 slices, each having 65,536 levels. A new Audio interface may record 96,000 slices a second at nearly 17 million levels for every slice. If you think that is a lot of data, well, you are right, it certainly is. The Bit Rate is usually expressed in Mbit/sec.

anonymous Fri, 02/23/2007 - 22:15

Bit depth

Well, I guess that sums it up pretty well. It's easy to get confused with all this bitrate/bitdepth stuff, especially when you're not really technically oriented like me. But I guess that in the end, it all comes down to what you hear. The thing is, you don't really hear your mixes until you've heard them about a hundred times, you know, until you can hear them in your head exactly the way it sounds on the speakers (speaking for myself, I ain't no pro). So some technical advice is allways good.
Thanx for the help JoeH. I'll be searching for your replies on this site. You really sound like you know your stuff.

Cheers man

Carlito

BobRogers Sat, 02/24/2007 - 06:16

Wikipedia has a good [[url=http://[/URL]="http://en.wikipedia…"]article[/]="http://en.wikipedia…"]article[/] on dithering with some examples that might help.

I had never heard the origin of the word before. If the article is correct, it was used to describe the way vibrations in airplanes made mechanical, analog computers more accurate by preventing stickiness (hysteresis) in the mechanical parts. Very cool.

anonymous Sat, 03/17/2007 - 14:08

Yes your right I realised my error as I wrote it. Dithering is bit depths anti alias filtering is for changing sample rates from higher to lower.
I always get confused with dithering and anti alias filtering which is what you apply when you change sample rates no ?
we used to be told that changing bit depths caused less problems than changing sample rates (unless the sample rate was exactly doublt the destination sample rate) - something to do with niquist .

dpd Sun, 03/18/2007 - 16:02

Digital processing temporarily increases the bit depth (multiplication of two 24 bit numbers produce a 48 bit result). These processes need to either dither or round the results (not the same) back to their output depth - this can happen internally many times depending on the algorithm's and programmer's requirements.

anonymous Sun, 03/18/2007 - 17:21

dpd wrote: Digital processing temporarily increases the bit depth (multiplication of two 24 bit numbers produce a 48 bit result). These processes need to either dither or round the results (not the same) back to their output depth - this can happen internally many times depending on the algorithm's and programmer's requirements.

Fascinating . Have you got a link to an article about this or perhaps a phrase I could search on Google , I have been unable to find anything about it. Are you perhaps referring to internal 48/32/24 etc bit floating point audio that most software uses ?

Cucco Sun, 03/18/2007 - 17:51

Guss -

I don't think you need to reference an article. Try your trusty calculator.

Multiply
n.nn
x
y.yy

you will come up with z.zzzz

Since you have increased the quantity of numbers after the decimal place, you have effectively increased the bit depth since only a finite quantity of numbers may exist after the decimal point given a specific bit rate.

Since sample rate conversion is a mathematical equation, performing any mathematical equation involving non-whole numbers is going to increase the quantity of numbers beyond the point, your sample-rate-converter has only two options (or a hybrid of the two).

Option 1 -
It can round to the nearest place (causing severe aliasing)

Option 2 -
It can operate at a bit rate higher than that to start with, work within that bit rate then dither back down upon completion of its equation.

It's a matter of simple math - no special article needed.

Cheers -

J.

anonymous Sun, 03/18/2007 - 18:14

Now hang on a minute I think I understand why Im getting confused here - Bitrate and Bitdepth are two different things no ?
As I understand it :

16 bit means that there are 16 binary units per sample (bit depth) and Bit Rate is how many bits of information are processed per second ( or unit of time).
The sampling rate is how many samples there are per second so obviously going from say 44.1 to 96 KHz would increase the "Bit Rate" because theres more samples and their component "bits" per second . But what I dont understand is why having more samples per second necessarily implies that each of those samples should have more binary units (bits).
Perhaps I need some sleep my head is swimming with strange half familiar phrases words, like Niquist, fourier and aliasing

anonymous Sun, 03/18/2007 - 18:15

Now hang on a minute I think I understand why Im getting confused here - Bitrate and Bitdepth are two different things no ?
As I understand it :

16 bit means that there are 16 binary units per sample (bit depth) and Bit Rate is how many bits of information are processed per second ( or unit of time).
The sampling rate is how many samples there are per second so obviously going from say 44.1 to 96 KHz would increase the "Bit Rate" because theres more samples and their component "bits" per second . But what I dont understand is why having more samples per second necessarily entails that each of those samples should have more binary units (bitdepth).

anonymous Tue, 03/20/2007 - 13:48

Michael Fossenkemper wrote: Guss, you're mixing things up here. Think of Bits as the amplituded and sample rate as the frequency. Bit Rate is generally used to describe streaming like MP3's. You are probably better off getting a book on digital audio as it would decribe the whole process much better than someone can tell you in a post.

Yes Michael you are quite right and I am reading a couple of good articles on the subject as we speak and I am beginning to get a taste of how complicate it gets . I hope you don't mind me discussing what I am reading as it relates to the subject.
Bit rate applies to all digital information not just mp3s , its just how many binary digits are processed per second ( for example a CD wav 44100 * 16 * 2 stereo = a bit rate of 1411 KB/S), sampling frequency is just how frequently a 16 bit sample is recorded (very similar to stills in a movie) and nothing to do with the actual frequencies of the sound recorded although if a frequency is higher than the half the value of the sampling frequency (niquist value) then it will not be represented though its subharmonics may be.
I still dont understand how the frequency information is represented by the bits if as eveyone says the each 16 bit sample only contains amplitude information, but I guess since frequency is a function of amplitude over time it must be related to that.
However I need to read more interploiation between sample rates and the extra bits neede because of thie you mentioned earlier and rounding etc
cheers,
Gus

dpd Fri, 03/23/2007 - 19:18

Gus - you are getting closer to understanding things. I strongly recommend you read Dan Lavry's excellent white paper on Sampling Theory: http://

Sampling rate defines the maximum frequency that can be captured - theoretically, it is one half of the sampling rate (Fs/2). In reality, due to the fact that real world filters don't have 'brickwall' characteristics, the highest frequency that can be captured is around Fs/2.25.

Cucco Fri, 03/23/2007 - 20:03

dpd wrote: Sampling rate defines the maximum frequency that can be captured - theoretically, it is one half of the sampling rate (Fs/2). In reality, due to the fact that real world filters don't have 'brickwall' characteristics, the highest frequency that can be captured is around Fs/2.25.

I know it's knit-picking semantics....but...
It's not the highest frequency that can be captured that is Fs/2.25, that's the highest reasonably-usable frequency portrayed at playback due to the filters.

In fact, the highest frequencies captured are technically just shy of Fs/2.

dpd Sat, 03/31/2007 - 21:45

^^^ Agreed (although I used Fs/2.25 as an example.)

Still, I'd hate to have to be the guy tasked with designing an anti-aliasing filter that has to get below 16 bit (or worse, 24 bit) rejection in a transition bandwidth of .5% - I'd be exceedingly concerned with the stage-by-stage stability of such a filter.

You know, I think I just figured it out - I'm talking about real-world high-rate analog filters in front of NON-oversampling A/D converters. With the oversampling converters (with an output rate decimated down to the 'real' Fs), I'm sure the highest frequency captured does get very close to Fs/2. Not in the former case, however - at least not alias-free, it doesn't.

Let's just call it a day, stick with theoretical limits and use Fs/2.