dead horse (the resurrection)

[stefancrs]

http://www.cockos.com/forum/showthread. ... 473&page=2

Just bits of it there, but more is coming soon, everyone's got AES right now

Only one post from Justin there, which I don't even get He says that errors in the 2 LSB's in 24bit audio results in errors when truncating to 16bit.

Or, no, I get it now. But it only applies if dithering isn't used. But even without dithering, the error wouldn't be a cumulative one as far as I can tell.

[stefancrs]

are you guys reading what you are writing?

It's not adding up the way it's supposed to............

What isn't?

[Pipelineaudio]

Only one post from Justin there, which I don't even get He says that errors in the 2 LSB's in 24bit audio results in errors when truncating to 16bit.

Or, no, I get it now. But it only applies if dithering isn't used. But even without dithering, the error wouldn't be a cumulative one as far as I can tell.

Ill get you more info in the next couple days. There was a pic from a test setup he did, but I can't find it...hate to sound like a lameass and maybe its not cumulative

But there should be more info in a few days

At the moment I am waiting in suspense to see if our friend here is just trolling or will actually put his bits where his mouth is

[MRoc]

the basic rules of addition for floating point numbers may be true for most cases, but...

32 bit IEEE floating point numbers have enough headroom for about 7 decimal digits, there are 23 bits for the mantissa. if you sum two signals that are different in level by, let's say more than these 7 decimal digits, the smaller one doesn't make really much to the sum. the FPU on PC adds both with much more than 32 bits and afterwards rounds them to to fit again into 32 bits when it goes back to memory, depending on the current FPU rounding mode.

it depends on how long variables stay in FPU with their higher precision and when they are put back to memory (and rounded). thus it is dependends on how code was written and what the compiler made out of it. that's why some compilers provide an option to prevent an optimisation of the order of FPU instructions cause this can have different results.

finally, there may be differences in the sum of different DAWS, but you probably need to sum really many tracks to meassure any difference and you probably won't hear that in a normal mixing situation.

here a ridiculous test:

take two 32bit floating point samples with each a sine of different frequency and a difference in amplitude of, let's say more than these 7 decimal digits. sum both several times in your favorite host, use a steep notch filter to remove the louder one of the sines. amplify the result by, lets say 130 dB. what should be left is the sine with the small amplitude. I'm sure you end up with something that isn't a sine anymore. what is left you can just call a FPU rounding noise leftover

best wishes,
matthias

[stefancrs]

MRoc, but this is a total non-issue, since it's a matter of SNR. Unless you have actual processes that add a ridiculously high-amplitude signal to a much smaller one and afterwards just removes the high-amplitude one. I don't see what the point of such a DSP process ever could be though, which is why it's a non-issue

[C00kie]

I'll tell you one that I do not believe [...] that I need to spend 99 bucks to solve a math problem. As far as I can tell, it's as much of a scam as anything......

1) this is not the only thing they changed in Live 7, just one icky picky detail they've fixed at the core. And at least they're being honest about it.

2) the only good reason to upgrade a host, is to get extra features you're going to use.

3) if this issue solely should convince you that Live has changed from a bad sounding to a good sounding host, than that's mental indeed. It was pretty good to start with, better than summing 16bit integers (which still does the trick for some engineers BTW)


I'll tell you something, I'm a software programmer for about 20 years now. Floating point arithmetics are a weird thing. You'll have to be meticulous in what order you do operations. A simple demo for you:

Code: Select all

<html>
<body>
<H1>Simple calculation rounding error demo</h1>
<script type="text/javascript">
var a = 101;
var b = 1.01;
var c = a + b;
document.write (a + " + " + b + " = " + c + "<br>");
var d = c - b;
var e = d - a;
document.write (c + " - " + b + " - " + a + " = " + e + "<br>");
var d = c - a;
var e = d - b;
document.write (c + " - " + a + " - " + b + " = " + e + "<br>");
</script>
</body>

Copy that into a .html file with notepad and run it in your browser. I get the following:

Simple calculation rounding error demo
101 + 1.01 = 102.01
102.01 - 1.01 - 101 = 0
102.01 - 101 - 1.01 = 5.10702591327572e-15

So I hope you see that it's easy to make tiny errors that won't get noticed easily.

[MRoc]

yes, stefancrs, that is a non-issue, that's why I call it a ridiculous test. but if people want to meassure something, they actually can (if they can)

[hibidy]

are you guys reading what you are writing?

It's not adding up the way it's supposed to............

Are you talking? Where's the test results?

my question exactly!

[stefancrs]

yes, stefancrs, that is a non-issue, that's why I call it a ridiculous test. but if people want to meassure something, they actually can (if they can)

hehe, yah.

[hibidy]

Only one post from Justin there, which I don't even get He says that errors in the 2 LSB's in 24bit audio results in errors when truncating to 16bit.

Or, no, I get it now. But it only applies if dithering isn't used. But even without dithering, the error wouldn't be a cumulative one as far as I can tell.

Ill get you more info in the next couple days. There was a pic from a test setup he did, but I can't find it...hate to sound like a lameass and maybe its not cumulative

But there should be more info in a few days

At the moment I am waiting in suspense to see if our friend here is just trolling or will actually put his bits where his mouth is

[stefancrs]

are you guys reading what you are writing?

It's not adding up the way it's supposed to............

Are you talking? Where's the test results?

my question exactly!

You're very cryptic. He meant tests that show audible differences. From you I think.

[Pipelineaudio]

my question exactly!

Download this test and do it yourself

http://www.kvraudio.com/forum/viewtopic.php?t=193425

If you dont try it but want to keep yakking in here, then you are just a troll stirring up shit, any other course of action makes you dishonest or a jackass...a jackass could be forgiven I suppose, but dishonesty? Thats just willfull

If you DO try it, and come up with results that dont null (except of course in the instances stated), then post them here and congratulations you will be a hero

Usually when you put frauds up to a real test they try and cop out saying shit like " oh I dont REALLY care that much" (even though they yakked for 4 pages)

Or "the test is rigged"

but I give you the benefit of the doubt

If you DO care, and arent just trolling, youll do the test

We will be here awaiting your results

[hibidy]

I don't have the resources to do one of these tests. Let's leave out the 99 dollar one since for some reason no one can understand what bugs me about it (er, at least no one who is posting)

The "thesis" is that all hosts sound the same.......right?

There is allegedly a test that proves this.......right?

So if ableton says.......and I'll quote directly

Enhanced Audio EngineLive 7's enhanced audio engine improves fidelity with precision 64-bit summing at all mix points throughout the program, POW-r dithering, optimized sample-rate conversion and other advances

How would someone come to the conclusion that THIS is a true statement?

I think it's a valid argument/question! But clearly to most of the current participants it is NOT a valid argument.....at least that can be agreed on

[hibidy]

You're very cryptic. He meant tests that show audible differences. From you I think.

I'm not making a claim!

I'm simply........again.......and AGAIN even......saying that if they are the same, and there is a test to prove it......how can live make a DIFFERENT claim.........

fewwwww.........this is just too hard.........come on guys.......I'm not the enemy here.......

[Pipelineaudio]

I don't have the resources to do one of these tests. Let's leave out the 99 dollar one since for some reason no one can understand what bugs me about it (er, at least no one who is posting)

The "thesis" is that all hosts sound the same.......right?

There is allegedly a test that proves this.......right?

So if ableton says.......and I'll quote directly

Enhanced Audio EngineLive 7's enhanced audio engine improves fidelity with precision 64-bit summing at all mix points throughout the program, POW-r dithering, optimized sample-rate conversion and other advances

How would someone come to the conclusion that THIS is a true statement?

I think it's a valid argument/question! But clearly to most of the current participants it is NOT a valid argument.....at least that can be agreed on

Tell me what you think "fidelity" means, and we will discuss it