Stereo Tool

That would explain it, I have everything turned on. And I'm using the maximum number of bands in ND and the multiband.

BUG FIX FOR CRASH IN BETA725-020.

Windows:
Winamp version: http://www.stereotool.com/download/dsp_ ... 25-021.exe
Windows stand alone: http://www.stereotool.com/download/ster ... 25-021.exe
Windows VST: http://www.stereotool.com/download/vst_ ... 25-021.dll


Changes:
- Limiter: Fix (check) limiter
- Limiting works MUCH better with no lookahead - seems to use the smoothed signal when using lookahead. Remove that and perform some smoothing in backward direction (probably based on drop speed) to remove crackle.

Under Detection there's now a "Limiter distortion" slider; setting it higher allows more distortion (which can be good for certain frequencies, for example bass).


(Copied from previous post) Description of Inertia:
So what is Inertia? There was a problem with the old compressors: After a very big volume spike, the speed at which the audio returned was always the same - but determined by how much it had to move up. So, if the volume dropped by 6 dB and after 100 ms the volume went up 3 dB, then for a volume drop of 12 dB that would be 6 dB. Sounds perfect.

But it's not. Say you have a huge drop, for example after a very loud 'S' in the high frequency band, where normal volume differences are at most a few dB and this S suddenly sticks out 20 dB.
For a difference of 4 dB, after 100 ms the difference in level is 1 dB - 75% of the difference is reduced. Now, this last 1 dB is really nearly unnoticeable, so for your ears the release kinda stops after 100 ms.
But, for a difference of 20 dB, after 100 ms the difference is still 5 dB! And you need more than another 100 ms before you reach this 1 dB point.

So, after a loud sound you hear a gap at settings that sound good for small volume differences.

The first thing that I did to improve this was to add Gamma (now called 'Flatness'). Which really helps at the final part of release: Small differences get compensated faster. But at the same time, bigger differences take longer to recover, which causes the same effect for really big differences as before.

Inertia fixes this. With inertia combined with Gamma you can make the release happen in a nearly constant time, without the slowdown at the end that you would have without Gamma, but also without the slower recovery for very big volume differences. Basically, the release happens in a nearly straight line, but the slope of the release depends on how much level must be compensated. With high Inertia values, release can even be faster for very big differences than for smaller ones, which can be good to quickly fill up the gap after a loud sound.

Final note: For bigger Gamma values you need bigger Inertia values. It might be possible to combine them into a single slider. But for now, there are 2.


Fixed previously:
- Added Release Inertia
- Improved limiter: Flatter frequency response, better behavior at low frequencies, more accurate level measurements, tighter (Bojcha, do you still hear the distortion from the previous beta? If so I can smooth things a bit more)
- Added Traditional Peak detection mode to compressor; this mode should not be used unless you have a lack of CPU power -> Brian, I think with this setting the new multiband compressor will perform about the same as the old one.
- Hangup when pressing Natural Dynamics in bottom display
- ND license incorrect
- AZIMUTH default settings: viewtopic.php?f=15&t=4875
- Phase rotator: Change name "As early as possible" (unclear!) and hide in most modes?
- Organize ND settings better
- Fix loading existing presets (Gamma values, "As early as possible" etc.)
- Split ND settings in to GUI modes
- Remove old ND implementation (saves memory and makes STS files smaller) -> 8 MB less memory usage
- Multiband: Improved Flat Frequency Response. Good values seem to be between 0.50 and 0.60 (depending on other settings!)
- Multiband: Added Attack and Release Gamma. See below.
- Phase rotator: Added checkbox to place it at the start of the processing chain (just before Natural Dynamics)
- Multiband/Natural Dynamics: Cleanup small spikes in Flat Frequency Response graph.
- Multiband/Natural Dynamics: Improve flat frequency response further.
- Mac: Default priority value to 46.
- Mac: Menu bar with Close APPLE+Q -> wxEntry exits but rest of software doesn't -> crashes.
- Mac: Cleanup named semaphores on exit
- Mac: RDS binary values
- Multiband/ND: Flat frequency response: value automatic conversion. ((v / .58) ^ .5) * .58
- Decide to which license to link ND -> Now linked to Declipper
- MB Knee does not work -> Yes it does...
- Linux: Test if everything still works. -> Yes, some small fixes were needed. Also removed printing to screen in Jack loop (bad!)
- Compressor: Place Gamma somewhere logical without making the interface too crowded.
- Display ND in bottom section
- ND handle changing number of bands
- Mac: Add auto-generating of .zip file with proper plist.
- Mac: Add icon
- Changed ND default settings (experimental!)

Remaining:
- Bob Hawkins built-in preset crashes
- Flat Frequency Response: Built-in presets use default value, should use conversion function.
- MB default frequencies on startup incorrect
- Default settings for ND -> Use latest from Bojcha based on JesseG's
- Fix Limiting display (must show dB's).
- Add presets
- Mac: VST: Simple workaround: Simple GUI with "Show Stereo Tool window" checkbox. Also has an (unplanned) advantage: Window is resizeable. Might even be a good idea to support this in Windows as well!

Later:
- ND artifacts band 1 (JesseG) - is this still an issue with larger drop times? -> Later
- Compressors: Add Reverse Release to improve start of loud highs -> Idea might be good but it doesn't work this easily. Postpone.
- Try to reduce MB/ND volume measurement CPU load. Maybe ignore some bands if there's nearly no signal there etc.
- Multiband/ND: Flat frequency response: Add separate settings for each crossover?)
- MB bands limiting display - Really difficult to fix (without increasing CPU load); it works fine without band coupling, and shows the actual reduction level if band coupling is used. So keeping it like this for now...
- Mac: VST: Child window support. Difficult... Maybe later!
- Mac: GUI performance.
- Mac: Cursors not supported. (Keep like that for now).
- AGC behaves bad: Different speeds at different locations.
- ND boosts highs in Dragana S track - should not do that-> later?
- Band coupling separate settings per crossover -> bit problematic, can this be postponed?
- MB Gamma and Inertia default values - but without changing existing presets
- Somehow try to estimate Attack/Release vs. Gamma/Inertia?

I see something strange nonflat - hump with using flat freq response slider in mb


10%

30%

_________________
Urban Junglist

Cannot reproduce it here - you're using 4 bands right? Can you send me your settings so I can see if something is wrong? Thanks....

I haven't really been around all that much to test things on my AMD system. I will tell you that the company laptop I have is a Core2 Duo L9400, and that runs things much better than the AMD, but my suspicion is still that performance on the AMD could be increased if the L1/L2 caches were streamlined. The way that you test filters in isolation, there is a good chance that the cache utilization is greater than what it needs to be, as you select the most optimized for each individual filter, which typically would be the implementation that uses the most memory.

At any rate, when I have a chance, I'll try and see.

Also, with respect to presets, again, in my opinion, people that are making FM or WEB presets should not automatically be downgraded into GENERIC. That whole section is extremely packed now, and it could be overwhelming to a user to have that many choices. Not only that, for any of us, like me, who have designed something specifically for generic, we get crowded out / down to the bottom portion.

Somewhere along the line, the delay has been reduced. With the current version, I am running the FM Output Buffer Size at .0007 seconds. Not long ago it was around .03 seconds. Using 512 Latency, I can easily listen to myself off the air monitor. It sounds like I'm in a barrel but it doesn't drive me crazy. And this is on a marginal system. That's a nice option to have.

If you use Phase rotator.. try to tune "low latency" settings there.
128 / 32 samples.

It was at 112 / 20. Haven't checked the barrel issue yet but it sounds more open with your settings. Thanks Bojcha!

I received the preset - the thin line is ok (peak is exactly at the configured frequency); the strange shape is caused by the big differences in steepness between bands. The thick line should then also be ok.

Changed defaults in next version

Note: The 2nd number (the 20 or 32) causes extra latency; you can make it bigger but every 44 (at 44.1 kHz or multiples thereof) or 48 (at 48 kHz or multiples thereof such as 192 kHz) samples is 1 ms of extra latency.