Stereo Tool

All this is because we start in wrong direction. But somehow this actually works.
I am sure that totally different techniques are used in other processors.

Current beta is actually best for now, and shure better then yesterday (hold time?) Less difference between hard compressed tracks and tracks with high rms/peek ratio.
I checked v5.0 and 4.22 versions, all actually do same thing .. lowers voices when it should not. That is why once asked for compressor after AGC, or kind of limiter or something, but again singleband compressor in ST is not actually real compressor, and long trip will be to make sidechain compressor or wb compressor .. or.. Once i ask for Final Limiter .. bands, ratio, lookahead...and still not sure how it works.
brrmm


Decay speed - Band 1=0.000251468
Attack speed - Band 1=0.000003805
Third band=0

Just thinking aloud here...

Basically, as long as all the sounds at all different frequencies stay within the same volume range, everything is fine.
Now if a single (or a few, in case of voices) frequency stands out - be it bass, a loud voice, whatever - we get in trouble.

For example, a loud voice in that Celine Dion track looks like this:

Here you can see why in this case the bass is actually helping to keep the volume constant!

Equalizing it according to ITU-1770 makes things even (a lot) worse:

The bass is lower, and the voice peak is higher, which makes things a lot worse.

Another troubling sound is Beyonce - Video Phone:


Which gets better when using ITU-1770 processing, but still contains an insane amount of bass.

That would lead to the following solution:
- I measure the frequency content of the input signal
- I find the 'normal' peak level (have to think of a calculation for that...)
- Using that, I calculate a 'maximum' peak level
- And then I kill everything that's louder than this maximum level (well, kill is too much. But I can reduce it by a number of dB's)

If I do that, I would process Video Phone without the bass. And Celine Dion without... Celine Dion
Most other sounds should not be affected at all.


But: This would lead to other weird things. For example, if I have a beep tone with other sounds, and everything but the beep disappears. Then I would start reducing the beep (for measurement purposes), and - in reality - boost its volume way too high. That could be solved by ONLY using this new level to determine if the volume should go down, and use the real levels to determine if it should go up.

But: Say there's only this beep tone, and the volume is already very high (the beep didn't cause it to drop). Then, when other sounds are added, suddenly the beep isn't dropped anymore and the total volume drops very strongly (because it didn't before). Yuck.

And if there's only a voice, the volume should be dropped if it gets loud.


Basically this is an unsolvable problem. I need to be able to distinguish "a very loud portion of this track starts to play, I need to very rapidly drop the volume" from "Oh, it's just Celine Dion, she'll shut up in a second, just ignore her". Without looking ahead (and the look-ahead time would have to be multiple seconds).


Still thinking... Ideas are welcome...



Edit: What would happen if band 1 does not respond to Celine Dion? So the volume below 200 Hz wouldn't drop if she starts to sing? Would that improve things? Just like band 2 is locked to never get lower than band 1, I could add a band where I only process bass and lock band 1 to that...

How about a very slow attack and a very slow release around the average input, with a faster attack and a faster release after that?

Hm... As simple as that sounds... It might just work... But they will need to be tuned to work together very precisely to make sure that a spike followed by a (relatively) rapid drop on the first doesn't leave a gap in the 2nd, and to avoid getting pumping at the start of a new track. For tracks with short spikes it should have a similar effect as my current shorter hold time solution, which might be an indication that this is the right direction.

On the other hand, there would still be a steep drop during loud voices, followed by a quick rise (which is a lot smaller than now though). That might be annoying, I'll have to test it to be sure.

I'll do some tests later - this is very simple to test by concatenating 2 Stereo Tools (in Winamp with DSP Stacker etc.)

Listened the newest version Can be... Stable and soft. OLL OK!!!!


Soundmanfm

Concerning the level at the beginning of song..For fm processors it is characteristic. It is possible.When music and DJs non stop intro soft ap it the normal!!! It admits. It is normal work Orban end Omnia. A slow reaction is important only without jerks. Megamix non stop it almost not notedly.

SoundmanFm

The following settings:
- First AGC very slow (down speed below .0001, up speed also a bit lower than default)
- Second AGC up speed a bit faster than now
- Output levels of both AGC's at the same level
This sounds better! (I have to go somewhere now, will do more tests later. But first results sound good!)

Just to add ...
Not only AGC is guilty for "voice-down", much more is because multiband...

I think multiband need some update too.

Separate multiband AGC and multiband limiter stages. Most other audio processors take a shortcut in the execution of multiband AGC and limiting functions; they put both functions within the same audio processing "block". Omnia takes a more comprehensive approach. We designed the multiband AGC and multiband limiting blocks completely separate from each other. This allows us to give slightly different treatment as needed for the absolute cleanest and competitively loudest audio. For example, some bands of limiting are best served by feed-back servo control, while the higher bands are best served by feed-forward servo control. Our design, which breaks these multiband functions into individual processing blocks, allows for the absolute best treatment of each audio band. ...

http://omniaaudio.com/omnia-6ex-and-6ex ... d-benefits

That's odd. It shouldn't - and I just checked and as far as I could hear it doesn't.

Have you set "limit output" to 100%? If not Multiband partially behaves as a singleband compressor/limiter!!! (I think I should just remove that setting, with the improved final limiter and loudness filters it's no longer needed, and it only damages the sound).


Note: Behavior of 2 AGC's is as follows:
AGC 1 uses default settings, but 'down speed' is near 0, and 'remove remaining peaks above' slider is set higher (200-250%). 'RMS smoothing' is turned off.
Result: The volume is dropped much more slowly. Songs with more dynamics come out louder (which is good!), and a loud voice that doesn't last too long doesn't cause huge volume drops (which is also good). The disadvantage is that new tracks are a lot louder at the start.

AGC 2 uses default settings, but the up speed is a bit higher. Remove remaining peaks above slider is set lower (150%). 'RMS smoothing' is turned off.
Result: Only the 'extra volume' that remains after AGC 1 is processed. The volume is dropped by at most 6 dB, but usually stays relatively constant. For loud voices, the volume is dropped, but it comes back rapidly (but only a few dB, which isn't as noticeable as before). Also, loud starts of tracks are lowered further by this separate processing step, which makes the output more constant; the lower 'remove remaining peaks' setting helps here too.

So I think this is the way to go.

I just did another A/B comparison (not completely honest, the two 'up speeds' combined sometimes cause the combined version to be upto 1.5 dB louder in silent areas of the song), but I did notice that for most of the loud voices, this new version has a louder output.


Edit: There's some bad news too. The worst case comparison (Gimme Gimme Gimme vs. a very hard compressed track) gave a difference of 3.15 dB before (1 band), now that's 4.40 dB. This might be a matter of tweaking the settings (increasing the up speed of the 2nd AGC and decreasing it in the 1st, after the 1st the difference is actually only 1.3 dB, the second is causing the problems), but it does require some further checks/tweaking.

(Note: 3.15 dB seems a lot, but with 2 bands version it's around 1.2 dB, and this is the worst case that I've seen so far).


Edit #2: Tried slower up time in 1st AGC, faster in 2nd.
For voices this works very well (the 1st has a very constant output level now, the 2nd removes remaining spikes at starts of tracks etc.).

Disadvantages:
- A lower volume track will start very soft, and remain soft for a longer time than before.
- There's a BUG in the AGC! I just found out that, when I boost the input volume of the 2nd AGC with these new settings by a factor 2, the difference in the worst case situation (Gimme Gimme Gimme) for 1 channel is decreased further by the 2nd AGC to 1.6 dB (!). However, if I *DON'T* boost the input volume, the 2nd AGC is actually *increasing* the difference by a few dB. I will have to check why this is.

Anyway, if that could be solved:
- The difference between tracks is reduced
- Effect of voices is reduced
- Softer portions of songs don't get played back a lot louder very rapidly - dunno if that's an advantage or a disadvantage though.

Note: Fixing the AGC issue that I found may change the effect on voices. Actually I expect it to improve further - loud peaks now cause the sound to drop too much...

Question,
1st band is "Full band" and 2nd is 200Hz and up. Did you try to put 1st band to 200Hz only.?

So .. if loud voice is at 500Hz and 2nd band is lowering that voice, then 1st band is lowering same too .. so actually double lowering.!?