Rendering audio files for less DSP usage?

[Philosonic]

Hi,

Considering those many threads I have read in this forum concerning problems of great DSP usage when Plugins such as 1176 or FC670 are applied in multiple counts, I am wondering if anyboby of you has ever experimented with rendering and reloading tracks (talking of Cubase SX).

I usually do kind of \"New York\" drum compression by creating a new file where only the drum tracks are kept, then process them with any plugins I want to (including bus compression usually with 1176LN), then export the stereo sum in 32 bit float format and finally import the stereo file into the original multitrack file.
I know that, in such a case, this technique works fine and is also often recommended but I would like to know if this procedure also makes sense in vocal- or guitar group procession. Is there a considerable detoriation of audio quality if digital summing is done in subsequent (but smaller) stages?

Thanx for opinions!

 

[Koed]

Have you ever tried using the 'Freeze' function in Cubase SX3?

(and yes. Consecutive summing will introduce more and more artifacts.)

 

[Akis]

Consecutive summing will introduce more and more artifacts.

Why? AFAIK, it shouldn't.

 

[svs95]

Consecutive summing will introduce more and more artifacts.

Why? AFAIK, it shouldn't.

Especially not if you do it at the native resolution of the DAW, i.e. 32-bit floating point in the case of Nuendo and many others.

svs95

 

[Koed]

Yes it does.

It's simple realy. Assuming that the original unsummed tracks are 24 bit, summing will quantise, alias and dither the tracks and sum them to a 32 bit file. This will introduce artifacts based on the original tracks.

The sum of those 32 bit summed files will always be different than the sum of all the original tracks. Because you are summing the artifacts too.

 

[svs95]

Yes it does.

It's simple realy. Assuming that the original unsummed tracks are 24 bit, summing will quantise, alias and dither the tracks and sum them to a 32 bit file. This will introduce artifacts based on the original tracks.

The sum of those 32 bit summed files will always be different than the sum of all the original tracks. Because you are summing the artifacts too.

By artifacts, I assume you mean rounding errors. At the level of a 32-bit floating point DAW engine with 180-bit accumulator, such errors will be at an order of magnitude less than the LSB of the 24-bit originals - which is ridiculously accurate, and there's no aliasing unless you change sample frequency (which a good SRC will transparently eliminate). It's why we use such tools.

So, Koed, what is your recommended alternative? Are you digiphobic? You think analog summing is a magic bullet? There are SO many truly significant ways to get bad sound in tracking and mixing that worrying about 32-bit floating point rounding errors is like worrying about what you're going to have for breakfast after a nuclear attack.


svs95

 

[Koed]

Well you were talking about Cubase SX which has a 32 bit engine.
I just recently switched to Sonar 6 with the 64 bit engine and the difference is not marginal but quite significant.

I'm absolutely no digiphobe. Been around computers and music since the Spectrum ZX81. I have also worked completely analogue and besides the 'hands on' thing of a real desk, there's no real quality improvement these days unless you're using a top of the line desk like SSL. I'm also not a fan of the outboard analogue summing craze. Some pieces of copper wire, some old transistors and transformer will give you the same effect. As will most of the convolution effects available now.

There's no real alternative for stem summing besides using a 64 bit mixing engine. You're right that with 64 bit the introduced artifacts are minimal.

Personaly I use the freeze function of cubase/sonar to conserve DSP/CPU usage. Everytime I'm happy with a track I'll freeze the effects.
This does not work on groups and I think this is a big omission in current DAW's.

 

[svs95]

Well you were talking about Cubase SX which has a 32 bit engine.
I just recently switched to Sonar 6 with the 64 bit engine and the difference is not marginal but quite significant.

I'm absolutely no digiphobe. Been around computers and music since the Spectrum ZX81. I have also worked completely analogue and besides the 'hands on' thing of a real desk, there's no real quality improvement these days unless you're using a top of the line desk like SSL. I'm also not a fan of the outboard analogue summing craze. Some pieces of copper wire, some old transistors and transformer will give you the same effect. As will most of the convolution effects available now.

There's no real alternative for stem summing besides using a 64 bit mixing engine. You're right that with 64 bit the introduced artifacts are minimal.

I didn't say that about 64-bit summing, I said it about 32-bit float, properly designed (as in SX and Nuendo) where rounding errors occur mostly at less than the 32nd bit, and occasionally at the 32nd bit, which is truly insignificant. It is why we use 32-bit floating point in the first place.

I agree that analog "summing" boxes are nothing more than "signal processors!" It isn't that analog summing is more precise and that digital somehow misses the mark (whether 32 or 64-bit). It's that in addition to summing, analog is adding various forms of euphonic distortion, phase modulation, harmonic distortion, and other inaccurasies that we like the sound of!

That being the case, there's no audible difference between summing through something like a Sumo (or an analog console) and running the stereo DAW mix through it :!:

So what this tells us is that if we want to achieve this kind of euphonic processing in a DAW, we need some ways to add harmonic distortion, phase modulation and perhaps more accurate analog tape kinds of distortion to the signal in addition to accurate summing.

Having said this, what we really need is not more accuracy in summing, but more characteristically analog kinds of inaccuracy. May I say to you that in this pursuit, 64-bit audio makes no sense whatsoever (aside from potential marketing value - "higher numbers are better, right?")!

Some of the latest analog emulation plug-ins (such as the UAD-1 Neve 33609) actually do produce the signal path distortions of the analog circuitry, and their Pultec does a fantastic job of phase modulation (both using physical circut modeling, not convolution).

Neve's Portico, and some others, have begun to offer serious tape emulation which is beginning to be available in plug-in form. Some of the latest plug-ins from Waves (such as the SSL line, and some new Neve emulations I've heard, but which are not out yet - but which sound incredible! actually even include (optionally) the analog noise of the original device).

So slowly but surely we are acquiring the means to "analogize" our digital mixes. The future, I believe, lies in that direction - not in carrying digital accuracy further than the LSB of a 32-bit float engine with 80-bit register and 180-bit accumulator (as in SX and Nuendo)! Lack of accuracy is NOT the "problem" with digital audio. Accuracy IS the problem. It's sterile and uninspiring.

Who'd have guessed that, back in the days when we were drowning in distortion and noise, and craving the accuracy digital promised to deliver? Now that it's here, we're beginning to realize we might have thrown the baby out with the bath water! Now we're trying to get it back, and I think we eventually will. But finally it will be under our control, and we can use it selectively, a little or a lot (or not at all) - which means it will all have been worth it.

As a friend of mine uses for his forum signature: "Distortion Is Art"


svs95

 

[Philosonic]

That´s what I thougt, svs95.
Actually I use Cubase SX2 which lacks major freeze functions. I haven´t yet upgraded and see no reason for it since it seems obvious for me that there is no difference between audio-exporting SINGLE tracks in 32 bit float and freezing them. The point is, does it make sense to render groups, e.g. guitar group procession/compression? If so, there is no reason to worry about DSP usage of UAD-Plugins (at least to some extent, of course you have to process most tracks alongside with each other). But if you ´ve got the feeling in which direction a voice, for instance, is to be directed by tweeking the cambridge, you could use the Neve 1073 in similar characteristics in a seperate project and render it - and there would be no further plugin instance count for voice procession!