DP's internal mix engine?

[steff3]

>> 32-bit floating point gives you a scaling 192db of dynamic range.

for how many channels of audio?

with 32 bit float adding ten times 0.1 (0.1 + 0.1 + 0.1 ....) is not 1.0 - so, headroom is one thing, rounding errors are the other - and without double precision .... mathematically ... well....

just my thoughts

best

[Shooshie]

I don't have time at the moment to read the whole thread; forgive me if I step on someone else's answers. DP has been based on 32 bits FLOATING POINT internal processing since as far back as I can remember. I do not know if the original version #1 was 32FP, but I do know that version 2 was.

When I asked a Waves engineer in the year 2000 why Pro Tools plugins were twice the price, I said "do they sound better than Native audio?" He replied, "Well, Digital Performer processes at 32 bits floating point internally, and that means you can't clip the audio internally. The potential processing power of 32 bits FP is enormous, and Pro Tools cannot do that, so DP is potentially capable of better and more accurate audio that Pro Tools." Yes, he really said that. He was an engineer, not one of their sales people.

So, DP's audio engine has always been top-notch.

That, of course, is unrelated to the 32 bit or 64 bit addressing system used by Apple and specified by the Cocoa programming environment and by OS X. The "bits" refer to entirely different things here.

What's important to know about 32 bits floating point vs. straight 24 bits, is that both process audio with the same resolution, but 32 bits FP uses w24 bits to resolve the sound pressure (dB, volume, loudness, whatever you want to call it) to one of millions of integer values (two to the 24th power) which leaves 8 bits left over for a very special purpose. Those 8 bits specify the starting range of any 24 bit value.

To use an analogy, let's imagine that a DAW processing audio internally at 24 bits is a submarine sitting on the ocean floor, and it has a window with a gauge etched into it, which you can use to look out into the ocean to allow you to see how high each wave sample comes up on your gauge. The gauge has millions of little marks on it for 24 bits of resolution, and it enables very accurate measurements of the waves (audio).

But 32 bits floating point DAW is a submarine with the exact same 24 bit gauge in a window overlooking that same ocean of waves so you can see those waves and measure them just like the other DAW, except it has one very different feature: the 32 bit FP submarine allows you to move that whole submarine up and down from top to the bottom of the ocean, maybe even flying above the ocean surface when needed. This allows you to place the window anywhere you need to see the details of that wave. So, if the wave has a huge spike, you can run that submarine up to the top of the spike and STILL measure it in 24 bits, gleaning all the details of that wave where it reverses its oscillation and comes back down to a zero crossing. Each sample specifies the "height" of the window with the little gauge marked on it in 24 bits, so you get to measure all that detail anywhere from top to bottom of the wave chamber.

In short, the 24 bit internally processing DAW has high resolution for audio, but it isn't much better at loud spikes than a 16 bit DAW. It just allows more accurate measurement within the ranges they can see.

32 bit FP can move its reference point up and down instantly, so it is virtually un-clippable. You almost cannot push it to overload, because it just sends that "window" up to where the action is and measures whatever is going on there with 24 bit accuracy. In the next sample it may send the same window down to the bottom and do the same thing in reverse. The advantages are almost infinite. I keep saying "almost," because everything has limits. Nothing we can perceive is infinite. It would take an awful lot of sound to distort 32 bitFP audio but if you said "never" someone would post an example tomorrow in which he had done exactly that. Probably not a real-world audio example, but one that would at least be a proof of concept.

Where our limits hit us hard is when outputting our 32FP mixes to 16 bit "CD quality" audio. Back to our submarine analogy, it's like going from the ocean of waves to an aquarium at Sea World. Of course it's better if you can stick to 24 bit output, but that's not the subject of this thread.

Shooshie

[David Polich]

I have Pro Tools 9 which I have to use from time to time as a "translation tool", either to
send something to a collaborator who is on PT, or to do a remix of something done in Pro Tools
and sent to me.

I don't know about PT versions after 9 but I can honestly say that PT9 Native sounds worse to me
than DP, by far. It has far less headroom than DP. Even with tracks panned hard left and hard right, somehow a mix sounds kinda mono. The Pro Tools workflow has always struck me as
bass-ackward, unintuitive.

The nail in the coffin for me, though, is Avid's lousy business model and nonexistent tech support.
So I couldn't care less whether more recent versions sound like a million bucks, I'll never give
Avid my money again.

[kassonica]

I have Pro Tools 9 which I have to use from time to time as a "translation tool", either to
send something to a collaborator who is on PT, or to do a remix of something done in Pro Tools
and sent to me.

I don't know about PT versions after 9 but I can honestly say that PT9 Native sounds worse to me
than DP, by far. It has far less headroom than DP. Even with tracks panned hard left and hard right, somehow a mix sounds kinda mono. The Pro Tools workflow has always struck me as
bass-ackward, unintuitive.

The nail in the coffin for me, though, is Avid's lousy business model and nonexistent tech support.
So I couldn't care less whether more recent versions sound like a million bucks, I'll never give
Avid my money again.

I agree about Avid, trust me.

But something changed from 10, using 64 bit floating instead of 48fixed for a start...

It sums (and really this is what the thread is about) and sum engine is very different to my ears especially when you bounce out a stereo file.

[Robert Randolph]

>> 32-bit floating point gives you a scaling 192db of dynamic range.

for how many channels of audio?

with 32 bit float adding ten times 0.1 (0.1 + 0.1 + 0.1 ....) is not 1.0 - so, headroom is one thing, rounding errors are the other - and without double precision .... mathematically ... well....

just my thoughts

best

Yes IEEE754 brings its share of wonky errors and tricky programming, however those errors occur well below the audible hearing range and below the johnson-nyquist noise that all analog gear is bound to.

The possibility, that despite all the other confounding factors, the OP was hearing the difference between a 32-bit float and a 64-bit float wide data bus is basically 0%.

Especially when the description of the difference between the two scenarios is an apt description of what pan laws do to the sound of a mix. (and of course that's not the only difference here!)

[Robert Randolph]

It sums (and really this is what the thread is about) and sum engine is very different to my ears especially when you bounce out a stereo file.

And once again why pan law is the most likely culprit.

Bouncing identical summed tracks to 24-bit int files in DP and PT (and every other DAW) has been proved to death to be bit for bit equivalent otherwise.

[mikehalloran]

As far as I can tell, PT 10 introduced 32 bit floating point processing in 2011, seven or more years after DP had it. PT 11 introduced 64bit FP in the mix stage only.

Does 32/64 bit floating point accomplish anything besides lowering the theoretical noise floor? I've never seen a technical paper that suggests it has another purpose. Most seem to suggest that 32 bit FP is already below the inherent noise introduced by the rest of the recording/processing chain.

[Phil O]

Does 32/64 bit floating point accomplish anything besides lowering the theoretical noise floor?

Well one thing is what Shoosie described in his submarine analogy. The resolution will remain the same over the entire dynamic range, whereas in a fixed point system as the numbers get smaller you start using fewer bits and resolution suffers.

Phil

[kassonica]

As far as I can tell, PT 10 introduced 32 bit floating point processing in 2011, seven or more years after DP had it. PT 11 introduced 64bit FP in the mix stage only.

Does 32/64 bit floating point accomplish anything besides lowering the theoretical noise floor? I've never seen a technical paper that suggests it has another purpose. Most seem to suggest that 32 bit FP is already below the inherent noise introduced by the rest of the recording/processing chain.

My understanding is the more the resolution, the less rounding and the LESS errors.

MagicD once confirmed this, every time you change gain, errors due to roundings are introduced, over medium to big documents they are magnified.

[bayswater]

As far as I can tell, PT 10 introduced 32 bit floating point processing in 2011, seven or more years after DP had it. PT 11 introduced 64bit FP in the mix stage only.

Does 32/64 bit floating point accomplish anything besides lowering the theoretical noise floor? I've never seen a technical paper that suggests it has another purpose. Most seem to suggest that 32 bit FP is already below the inherent noise introduced by the rest of the recording/processing chain.

My understanding is the more the resolution, the less rounding and the LESS errors.

MagicD once confirmed this, every time you change gain, errors due to roundings are introduced, over medium to big documents they are magnified.

From a practical perspective, that's the same as lowering the noise floor (or increasing head room).

[Robert Randolph]

As far as I can tell, PT 10 introduced 32 bit floating point processing in 2011, seven or more years after DP had it. PT 11 introduced 64bit FP in the mix stage only.

Does 32/64 bit floating point accomplish anything besides lowering the theoretical noise floor? I've never seen a technical paper that suggests it has another purpose. Most seem to suggest that 32 bit FP is already below the inherent noise introduced by the rest of the recording/processing chain.

My understanding is the more the resolution, the less rounding and the LESS errors.

MagicD once confirmed this, every time you change gain, errors due to roundings are introduced, over medium to big documents they are magnified.

Those errors are happening at such an infinitesimal level though. Below -190dbFS, that's 70db below the noise floor of your converters most likely.

This is a 'forest for the trees' sort of thing. There are so many other confounding variables that are making a clear, intended and high level change.

[waxman]

The cat's STATEMENT was ProTools mixing engine now is "NOT" collapsing during ITB mixes. The cat's statement can be interpreted and theoretically challenged a gazzillion ways ending with a thread of 32 or even 64 pages.

After all the "barking, submarines, magic quotes, version numbers, equations, analogies and spin..." For sake of clarity and the simple mind may I suggest 3 interpretations while this tread is still in single digits?

Does ProTools in the box mixing engine sound BETTER, WORSE, SAME as DP 32bitfp ITB?

After all the sound of the final mix is an important piece of the puzzle agreed?

[kassonica]

The cat's STATEMENT was ProTools mixing engine now is "NOT" collapsing during ITB mixes. The cat's statement can be interpreted and theoretically challenged a gazzillion ways ending with a thread of 32 or even 64 pages.

After all the "barking, submarines, magic quotes, version numbers, equations, analogies and spin..." For sake of clarity and the simple mind may I suggest 3 interpretations while this tread is still in single digits?

Does ProTools in the box mixing engine sound BETTER, WORSE, SAME as DP 32bitfp ITB?

After all the sound of the final mix is an important piece of the puzzle agreed?

Interesting that you should post this, as I'm going to grab the consolidated files from a song from that session, and bounce out the files in DP and listen and check, in this test I will pan everything in mono to negate pan laws..

Ps I am NOT the only one who has noticed the mix engine has changed in PT..

if the files null, then there is NO difference...

[mesayre]

If you're noticing it specifically when bouncing out files at lower resolution, perhaps it's a difference in dithering. As that would be most evident on decays and low-level sounds, I imagine that could cause a difference in perceived depth. PT lets you choose from a few.

Maybe try throwing DP's Quan Jr. Plugin on there and see if that gets you any audible difference.

[kassonica]

If you're noticing it specifically when bouncing out files at lower resolution, perhaps it's a difference in dithering. As that would be most evident on decays and low-level sounds, I imagine that could cause a difference in perceived depth. PT lets you choose from a few.

Maybe try throwing DP's Quan Jr. Plugin on there and see if that gets you any audible difference.

Nope bounced the files out at the resolution they were recorded at 96/24...