Audio Engine?

[AD80]

And even then, how much does it really matter? I've yet to hear any software which sounded so badly that I wouldn't use it. That includes Ejay Ibiza! It sounds great. I don't have a golden ear, but I'd bet that my ear is a little more focused than the average music consumer. A great deal can be wrong with a song and people who don't make music won't notice it no matter how many times they listen to the same song. They will never notice it, even when the singer is off by a mile. In other words, if these hosts do sound different, I don't think it matters. They all sound good enough to make a hit record and people are making excellent quality music on all of them.

Right on!!

[monsterbeetle]

I'm OK to say it's the dumbest thread ever, but what's putting me off is when people start to mix up several uncorrelated problems together to produce inconsistent arguments, such as :

subjective quality of musical composition mixed with subjective workflow appreciation mixed with audio summing mixed with 32 bits vs 64 bits effects mixed with VSTis etc... when talking about the quality of an audio engine...

We can talk uninteresting or geeky stuff, but should at least do so correctly.

I'm off

[AD80]

I dont even know why everybody's arguing. All the pro's know Tracktion has the best sounding audio engine on earth at the moment .

[Spe3D]

[kilroy]

"If host A sounds better than host B because host B has bugs and host A doesn't, then we are on to something."

Another way to look at it would be...

If host A sounds better than host B, but crashes more than host B, then Host B sounds better than host A.

[Mr. Slater's Parrot]

If host A sounds better than host B, but crashes more than host B, then Host B sounds better than host A.

Well, certainly, no one likes it when an application crashes.

The original point, however, was focused more around the likelihood of two hosts computing different numerical results when mixing/summing the exact same audio sample values when both hosts are using 32-bit floating point.

[Phaedo]

I find that Magix Music Studio sounds much better than Cubase SX if I've got the speakers switched on for MMS and switched off for Cubase.

[VitaminD]

ooh looks like I missed this thread.. by a year


I guess it was for the best but I was still fairly well interested in the tests.. so were they posted?

[annode]

I actually read all 7 pages just now...and as far as some saying it`s "the dumbest thread of all time"...I did benefit from it.

I`m running nuendo,and have a habit of liking to use lots of vsti/vst.
As a project gets very involved,(approaching 50% cpu workload),the sound begins going to hell and I had assumed it was just the high cpu level only.

Now,after reading this thread,I understand more about what`s happening.
Luckily,I can make appropriate host settings and change my work techniques to get the best sound possible for different sized projects.

So it seems,more then one work technique will be needed, considering the cpu load.

I`m happy to understand,from reading this thread,how I work and have my host settings are the reasons for my sound degredation and not my software/hardware.(pc/adat)

[snareSpanker]

The way audio summing is handled is indeed a real problem,

for instance to add two 24bit words you need in the extreme case a 25bit word to represent the results see the problem? So the guys used 32 bits float, where you have 1 sign bit , 23 bits of mantissa, and 8 bytes for the exponent. So okay, the dynamic range is greater thanks to the exponentiation, but the mantissa is smaller... see the problem again? with a decent number of tracks you can exhibit an alterations on the sum, because when using floating point numbers, the quantifications noise level is correlated with the amplitude of the signal, whereas it's not with fixed point numbers.

the other factor is that sound cards can't output 32bit flows, so the signal is dithered, and differently from app to app. for reproductibility, export the mix as 32bit floating point, and listen to it in the same host, with the same dithering, to be able to isolate the impact of the summing.

The floating point numbers do not behave as their perfect mathematical counterpart in R do. However there have been some tryouts at implementing real numbers packages, using another representation of the numbers, different from the mantissa+exponent way (this time a real number is a size varying list of its integer part plus irreductible fractions) arithmetic operations have been defined on these numbers, and those numbers behave in a mathematically correct way. But the computations such as multiplication, division, and even additions take just ages to complete. However with these numbers, you have no rounding error at all, no noise induced by truncatures, etc... they are genuine real numbers

Would be nice to have a "High quality export" module based on this implementation of real numbers. but it would take for instance a week to export and dither a 40 tracks project.

For those interested in Comp-Sci, just take a look at that:

http://citeseer.nj.nec.com/375625.html

and stop dickin' around. The audio summing problem is REAL.

@vitaminD

Thank you for bumping this thread. After reading it all (..) I finally understood how stupid I am. Thanks alot again