A few "Dither" and CD burning questions (sampling and Bit-Rate)

JohnnyAmato

New member
Been doing a lot of research on this, and sorry if it's still a little confusing to me.

I'm making rough "masters" of my material for listening in the car, uploading to my music sites, etc.

The bulk of my material is recorded at 24/48 in Pro Tools 12, and if I want to burn a CD for referencing on other systems like the car, I use Sound Forge 10 to resample to 44.1, then burn (I assume Sound Forge's burner itself changes the Bit-Rate as it burns, as all I do prior is change the sampling. Uploading to sites, I simply upload the 24/48 WAVs.

My questions are these:

For quick rough masters, I'm using the Slate FG-X on my master bus, which has a Dither option. If I'm bouncing out at 24 bits anyway, do I still need to dither? Also, I do my final fades to the bounced master in Sound Forge, and intro and ending cuts (silence), which technically is more processing. Should I not have dithered in PT yet then? Sound Forge has a Dither option also, but only if you change the Bit-Rate manually, which I do not. I only change the sampling rate down to 44.1, which that operation has no Dither option. I just would like to know if I'm doing things right. I suppose if I got more used to doing the final fades in PT itself, it might make things easier, but I still would need to know when to Dither. I've read many conflicting arguments in different threads, and would like to see what people around here think.

Does a 24-bit song/file uploaded to a music site need dithering at all, or is it just needed for conversions down to 16-bit for CD?

The weird thing is, by changing the sampling rate in Sound Forge down to 44.1, it now lets me burn CDs, assuming it automatically changes the Bit-Depth to 16 during the burning process. But if I just change the Bit-Depth manually down to 16, the sampling is still 48,000, and won't let me burn the CD. Again the Bit-Depth operation is the only place I have a Dither option in Sound Forge, so that's adding to my confusion. Should I be doing both of theses steps manually, and adding dither at this final stage, instead of in PT on my master bus?

I hope this all made sense, and any input from the knowledgeable would be GREATLY appreciated!
 
Generally speaking, truncate/dither should be the very last thing that happens to the audio. But for checking mixes rather than release it's not absolutely necessary.

When you truncate (e.g. from 24 to 16 bit) it leaves a ragged "edge" that can interact with the signal in unpleasant ways, so a type of very low level noise is added to randomize the edge and prevent the interaction. Mostly you won't hear the problem except when your signal is low, like during a fade out.

I like to dither/truncate as a separate, final, process if I'm not certain the software is doing it in the right order.
 
For PCM audio for music, dither is a fix for a mechanical flaw with how the system works. Every "bit" represents 6 dB dynamic range. The loudest sound you can have uses all of them. Within the range of the very first bit is where the sound starts from. Less than one bit of signal is the zero crossing of the waveform. No bits means no signal. Silence.

The very first jump in signal creates a problem. Using 16 bit as an example, the theoretical range is 96 dB. The first bit can handle either nothing, or -90 dBfs. So if you put a -93 signal into the converter it gets rounded to -90. This creates distortion. This kind of distortion is about as bad as it gets. It's not smooth tube amp distortion. It sounds horrid. It's called truncation.

As you add more bits to the signal (meaning as the signal gets louder) the range of values that can be represented gets much better and the distortion problem doesn't happen. Understand that where this distortion happens is in gnat fart territory. You aren't likely to hear it directly unless you amplify it a lot. The consequence it has is easier to hear if you're trying to evaluate the emotive characteristics of your music. The 3D depth perception potential gets flattened. Things turn mushy. Reverb tails get choked.

If you're working at much more normal levels that are far louder than the zero crossing of the waveform, it doesn't matter. All of your waveforms will have to cross zero several thousand times per second. It's not a problem as long as there's no samples that actually land in there causing the problem. There's a certain (eg. 100%) amount of likelyhood that it will happen.

Dither is a device that makes the very first bit fluctuate. Instead of being controlled by the signal, it gets controlled by a random probability generator. It makes noise, very much like static. Still in gnat fart territory. In order to mask the distortion caused by truncating the signal, the level of noise you would need is much greater than what dither puts on the signal. By randomizing the least significant bit and introducing broadband noise, the distortion gets killed or removed or more accurately, decorrelated - not masked. It gets trapped inside the dither. You would hear the -93 dB signal with noise around it, but no distortion. The reverb tails come back. The distortion is caused by and very much related to your signal. Dither decorrelates that into broadband noise that the ear isn't really drawn to. At an overal quieter level. As a separate entity to your signal.

Dither needs to be applied whenever the signal gets quantized. As your signal goes from analog to digital, the converter should dither it. A lot of them probably don't even give you a choice, they just do it. If you put the signal into a DAW, they don't process it in the same way. Most DAW software will work in floating point math. That means if you select 24 bit recording, it's a fixed binary word handling signal level. "32 bit mixing" should usually mean 32 bit floating point. It's exactly the same as a 24 bit fixed point word, with an 8 bit exponent added to it that can slide the range and resolution of the signal around as needed. Nobody has figured out a way to dither floating point audio. Floating point audio doesn't really truncate anything anyway, so we don't need to dither it.

The signal will then go to your DA converter using fixed point 24 bit and on to your speakers. So if you record your signal and put it into a DAW and don't change anything, you can export it without dither and no truncation will happen. If you change the level by exact increments of 6.02 dB, dither is not required because the samples will line up in approximately the same relative spots after they moved. No truncation. If you change a level by 1 dB, or any increment that isn't 6.02, you're essentially forcing the audio to be requantized. The samples will move and truncation will happen. For this reason, a dither plugin as the last thing on the master buss in your DAW will basically kill any truncation.

It's more critical and much easier to hear when working with 16 bit files. In 24 bit, a lot of this stuff is happening at theoretical signal levels that are out of range of anything and everything in the signal chain. When you render a mix, you're writing that stuff in it. You can't take a truncated file that's been written and put noise or dither or whatever on it and kill the truncation. you'll just have noisier truncation. So for 24 bit files, even though it's not likely to sound different one way or the other, dither is basically going to save you in the event of subsequent processing that could add more truncation. It's cumulative unless you kill it.

Pretty much every delivery format these days is subject to subsequent processing. If somebody converts a 24 bit file that you uploaded, and it truncates because it wasn't dithered on export there's nothing you can do at that point. Forget about the guy who made an MP3 out of the M4A that he ripped off of the internet. Could be swarming space goblins by that time.

Noise shaping is another thing related to dither. It allows you the choice to move the noise of the dither out of the midrange band where it's easiest to hear. Noise shaping should only apply in the last stage of processing. Since we don't have a last stage of processing anymore, it's probably best to avoid that altogether. It can hype up the high frequency energy of your stuff in a way that doesn't sound all that good.

When in doubt, just throw a dither plugin at the end of the master buss and forget about it.
 
Thanks a bunch, guys. I'm definitely getting a better understanding. It seems if I'm going to be doing my fades and conversions in Sound Forge, I'm better off not dithering until then, at the very end.

So after limiting on my master bus in PT without dither, I would then open the song in Sound Forge, do any final fades and trimming first. Then resample to 44.1, then change the bit depth to 16 with and apply dither then. And this is only if I'm going to burn a cd. If I'm just uploading 24/48 songs to my music sites for listening purposes, I won't bother with dither at all.

When the time does come for professional mastering for distribution, I'll obviously be leaving all limiting off the master bus, and won't be dithering at all, I'll let the mastering engineer deal with it. This was all just for my own listening/uploading purposes, and wasn't sure how important dithering was for that. The cd's I'm burning are only for myself for checking mixes on other systems. And anything I upload to my music sites are not available for download yet, until they are professionally mastered.

Thanks again for the help :guitar:
 
Just nit-picking: In this context, you are talking about bit DEPTH.

Bit rate is how many bits per second are being used, which is how mp3's are generally measured because some have a variable bit depth.

If something is 44.1k 16 bit, it has a bit depth of 16 bits and a bit rate of 1411kbps. (44.1k * 16 * 2 = bit rate of a stereo 44.1k 16 bit file)
 
I'd be willing to bet you wouldn't hear the difference unless you went searching for it. I'd be more worried about the sample rate conversion myself.

That's not a reason not to dither! It's free, slap it on there. Heck, it's only a tiny bit of noise anyway, add it any time you want*. But especially for "quick and dirty", like you could have done it and been rocking down the road by now and I promise you wouldn't hear the quantization distortion.


*I usually add more noise than this to full album masters anyway.
 
ashcat_lt said:
I'd be willing to bet you wouldn't hear the difference unless you went searching for it.

Once you know what to listen for it becomes a lot more difficult to not hear it. Bob Ohlsson is a mastering engineer that has had clients send him their stuff. A few times, he's gone back to them to ask for another print of the 24 bit mix with dither on it. Every time this has happened, there was no dither on the first print. The source plays a role here as well. He admits he's probably done a bunch of other stuff where he couldn't tell whether or not it was dithered. 16 bit is much more obvious.


ashcat_lt said:
It's free, slap it on there. Heck, it's only a tiny bit of noise anyway, add it any time you want.

It's a much tinier bit of noise than any tape machine ever. Plain old triangular or TPDF dither can be applied multiple times without changing the character of the mix. The consequence of using TPDF dither is always going to be far less than the potential consequence of not using it. Flat, hard, mushy, dead sounding "digital crunchy" isn't doing you any favours.
 
[MENTION=40104]snow lizard[/MENTION] - I wasn't taking to Bob Ohlsson, I was talking to JohnnyAmato. ;)

I don't think we're actually arguing.
 
I wasn't really trying to argue anything, just putting up information that I've learned on the subject. Don't get your knickers in a twist dude.
 
The timing of dithering has always been perplexing to me as well. I record in Logic at the conventional 24 bit / 44.1 kHz. I've been careful to avoid 'premature' dithering from certain mastering plugins, like Waves Ultramaximizers, and I have to manually turn that off in those plugins. The reason is because, like Johnny, I routinely burn CDs to audition my semi-mastered mixes in my car, and I also may want to upload/deliver the non-dithered 24/44.1 output (master) mix to various forums. When Logic burns a CD, the bit depth is automatically reduced to 16 (or course), and dithering is automatically applied, unless you get into the higher level controls on Logic's burning process and actively turn that off (and there is also an option for selecting the dithering algorithm). This is not apparent from the basic Bounce window that pops up when I hit the Bounce button on the Output channel. Since I had heard (repeatedly) that redundant dithering is undesirable (for the reasons explicitly outlined by Snow Lizard), I just simply avoid all dithering until the final burn to CD, and I let Logic's default burn processor handle that.
 
For clarification on when to dither, my advice would be any time the signal gets quantized. Another way to think about it is any time the word length gets reduced. It's more critical with 16 bit files, so absolutely if you record and mix a project in 24 bit and then run it down to 16 bit for a delivery format, dither should be used.

If you want to do a mix and then have it sent for mastering or use a separate program for further processing, there are a couple of scenarios. If you record at 24 bit, do a mix at 32 bit float and then export it as 24 bit fixed, dither is recommended. Yes, you would have to dither again at the next stage, but exporting to a fixed point format will truncate, and dither will get rid of the distortion. If you choose not to dither at this stage, the truncation distortion gets written into the file as signal, and can't be removed after the fact. If you can't hear the truncation at that point you're certainly not going to hear the dither. Any truncation present at that point might not be apparent until further processing is applied. It's too late to do anything about it then.

If you do a mix in 32 bit float, and save the files as 32 bit float, dither is not required.

Dither is like Molly Maid. It keeps your signal clean.

I personally do not recommend noise shaping.
 
Back
Top