Stereo Tool

Phoenix, think of it like this.

a = 11
b = 2
c = b / a = 0.18181818181818181818.......
Rounding to 2 digits gives c = 0.18
Now c * a = 11 * .18 = 1.98.

So depending on how smart the compiler is, the following code can give different results:

c = b / a
d = c * a

If a compiler is dumb, you'll get c = .18, d = .18 * 11 = 1.98.
If the compiler is smart, it will see that d = (b/a)*a = d = 2.

And if the compiler is strict, it will still reply with 1.98 because that's what the programmer asked - this makes floating point calculation optimizations nearly impossible (and I've turned this strictness off).

Moving code around leads to different "smartness" of the compiler, hence to (slightly) different results.

This keeps happening, the only thing is that I should rectify the slider is moved to another preset value of this and 20% (depending on initial state)

_________________
by GAP
"Less is More" (Bob Katz)

http://mathworld.wolfram.com/RoundoffError.html
http://www-aix.gsi.de/~giese/swr/allehtml.html

Lol! Some great examples and stories there

BETA032:
- Problem with lower Quality setting should be fixed now
- Overall audio quality should be improved (biggest difference is at low latencies): There are less artifacts now (no CPU load effect, but the audio is closer to the optimal audio at the double CPU load). *MAYBE* the voice vibrations are reduced as well. At lower Quality settings the artifacts might actually be worse than before, although I've also fixed a bug in the handling of lower Qualiy settings that caused artifacts. I may need to look into this further.
- Slightly (a few hundred kB I think) reduced the memory usage.

Winamp/DSP: http://www.stereotool.com/download/dsp_ ... 22-032.exe
Stand alone: http://www.stereotool.com/download/ster ... 22-032.exe
VST: http://www.stereotool.com/download/vst_ ... 22-032.dll
Command line: http://www.stereotool.com/download/ster ... 22-032.exe


IMPORTANT: LOW LATENCY OUTPUT IN THE STAND ALONE VERSION IS BROKEN AND MIGHT EVEN CAUSE CRASHES. SO MAKE SURE TO TURN IT OFF BEFORE UPGRADING! The good news here is that with the new design it should be pretty simple to get a much better working low latency output which might even have a lower latency than before.

I have received a report about sudden deep volume drops in BS412 (once or twice a day), I'll look into that tomorrow.

Ah ... clear now. Thanks for exemplifying this. So, if I understood you right, post organizing your code, there's a variance in 'obscurity' to the compiler for such floating point operations. Nothing relevant to the output of the filter quality.

_________________
visit website

Yes. At least that's what I think it is. I compared the audio of the latest build (before the quality improvement) against that of 029 and I saw some difference in the last bit, sometimes even the last 2-3 bits, in some presets (some others were 100% identical). Since I didn't change anything in what the code should do, and real bugs would cause *much* bigger differences, I figured it has to be a rounding issue.

By the way, if I turn *all* the optimizations off, the difference can sometimes even amount to over 5% of the total signal! Which is quite insane... (And runs at about 1/10th of the speed, so it's completely unusable).

Hm. Bojcha ran some tests and voice vibrations are not changed. But... His analysis did show what is wrong in the current clipper protection (if you set the bass clipping to -6 dB the vibrations are gone, BUT far too much bass is gone too. And that SEEMS to be caused by a wrong filter order).

Could you notate a specific track that has clearly audible voice vibrations, not something determined by scopes or frequency analyzers?

The reason I ask is because I was told that my H7RC1 had vibrations with a specific track, and I flat out do not hear them. I have posted H7RC2, and had asked that it be checked to see if the alleged situation was better or worse, and I have not been given a reply. As such, I am about to finalize H7RC3, which will probably be the final RC unless someone can demonstrate to me the alleged issue.

Thanks...

The worst track I know is one that Bojcha sent to me over a year ago, http://www6.zippyshare.com/v/35161681/file.html . The worst part is the "woo" at 0:14.