Stereo Tool

It`s been optimized with the newest protocols and this is the result . Old ones below K10 are getting almost useless

Latest version (022)?? CPU load went up a LOT a few versions ago but now it should be lower than it has ever been (but I haven't really tested lower latencies yet). Only if you enable "Slow FIR (test)" (Extreme Tweaker mode) then the CPU load will go through the roof...

Edit: Measured it here, and at all latencies the reported CPU load in TaskManager is 0%... And the sound is what it should be. Can you tell me exactly what's happening? And really make sure that you have the latest build, previous builds contained errors! (The new one has a range of 60 ms).

W.r.t. band coupling/linking, some more info which should help to remove bad options.

Right now there are 4 options:
1 Merging, % (current method for linking)
2 Merging, dB's (current method for coupling)
3 Minimum, %
4 Minimum, dB's

Let's assume 50% band coupling from band 2 to band 1, band 2 is 12 dB more reduced than band 1 without coupling or linking. Now regardless of the level, the relative levels of band 1 and 2 should always remain equal.

So let's examine 2 situations:
- Band 1 at 0 dB, band 2 at -12 dB
- Band 1 at -12 dB, band 2 at -24 dB

Attenuation: 0 dB = 1, -12 dB = 1/4, -24 dB = 1/16.

1 Merging, % (current method for linking)
b1_out = min(b1, b2) * link + b1 * (1 - link)
= min(1, .25) * .5 + 1 * .5 = .625 (for 0, -12).
or min(.25, .0625) * .5 + .25 * .5 = .15625. (for -12, -24)
.15625 * 4 = .625. So this result is EXACTLY 12 dB lower, which is exactly what I want. So this method is ok.

2 Merging, dB's (current method for coupling)
b1_out = 1 / (max(1/b1, 1/b2) * link + 1/b1 * (1 - link))
= 1 / (max(1, 4) * .5 + 1 * .5) = 1 / 2.5 = 0.4
or 1 / (max(4, 16) * .5 + 4 * .5) = 1 / 10 = 0.1
0.1 * 4 = 0.4. So this again results in EXACTLY 12 dB difference. Good.

Btw, .625 corresponds with about -4/-5 dB, 0.4 with about -7/-8 dB. Method 1 and 2 are both ok, the difference (as mentioned earlier, and shown here again) is that method 2 leads to more attenuation if the reduction is big. Will test it below with small values.

Method 3
link_from_band_1 = 1 - (1 - b2) * link
= 1 - (1 - .25) * .5 = 1 - .75*.5 = .625
or 1 - (1 - .0625) * .5 = 1 - .9375 * .5 = .53125
b1_out = min(b1, link_from_band_1)
= min(1, .625) = .625
or min(.25, .53125) = .25
As I wrote earlier, a much deeper band 2 has less effect on band 1. So this version is WRONG.

Method 4
link_from_band_1 = dBs * link
= -6 dB (for b2=-12)
or -12 dB (for b2=-24)
-->
min(0, -6 dB) = -6
or min(-12, -12 dB) = -12
WRONG AGAIN. The difference should be 12 dB, not 6.

Conclusion: Both methods that I'm using now (the one for linking and the one for coupling) are fine, both other methods will fail.


Now for the difference between method 1 and 2. Let's try it again with b1 = 0 and b2 = -3 dB (.707) instead of -12.

1 Merging, % (current method for linking)
b1_out = min(b1, b2) * link + b1 * (1 - link)
= min(1, .707) * .5 + 1 * .5 = .8535

2 Merging, dB's (current method for coupling)
b1_out = 1 / (max(1/b1, 1/b2) * link + 1/b1 * (1 - link))
= 1 / (max(1, 1.41) * .5 + 1 * .5) = 1 / 1.205 = 0.8299

Both values are close to -1.5/-2 dB. So where with a difference of 12 dB the effect in dB is almost doubled in method 2 compared to method 1, for small differences it is not! Method 2 still gives more reduction, but the difference is much smaller. So for very big differences, method 2 gives relatively more effect than method 1. Which was to be expected, and corresponds with what I wrote earlier a few posts ago.


Conclusion: The only thing I need to do is check if I should use method 1 or 2.
Method 3 and 4 are bad.


What remains is this question:
If a band goes down VERY deep, should it pull adjacent bands down equally hard (method 2) or should it pull less hard (method 1)?

In the current implementation, Band Coupling (with the matrix) uses method 2. Band Linking (3->2, 2->1 etc.) uses method 1.


Finally, there's an extra method that has not yet been considered: Coupling a band a constant number of dB's above the level of another band.
Formula: b1_out = min(b1, b2 / link)
So for example,coupling band 1 to 6 dB above band 2 would give:
0, -12 -> -6
-12, -24 -> -18
Which is a 12 dB difference. So this method would also be good.


FINAL CONCLUSION.
There are 3 possible methods that give 'good' results:
1. b1_out = min(b1, b2) * link + b1 * (1 - link)
Effect gets smaller if the relative difference between bands is bigger.
2. b1_out = 1 / (max(1/b1, 1/b2) * link + 1/b1 * (1 - link))
Effect stays the same if the relative difference between bands is bigger.
5. b1_out = min(b1, b2 / link)
Effect is constant at a margin above band 2, only if band 1 gets too loud the level will be reduced further - no smoothing area as in methods 1 and 2.


As said, for Band Coupling I'm now using method 2, and I'm starting to think that I should do the same for Band Linking. Method 2 is closer to Method 3 than Method 1.

It's what I'm using now! But this is the furthest away from what I think you asked, which is method 3 (linking would stop a band from going up more than a number of dB's above the level of the source band)

Let's try this with reasoning first. I'll implement it tonight so we can actually test it tomorrow

Let's for now just look at the 2 highest bands with band linking (the new thing). Say we have a 7-band preset, then we have linking 6->7. Let's say 50%. (At 0% and 100% all methods behave exactly the same!)

Now consider the following situation: The music contains a loud S sound which doesn't contain much audio in band 7. Band 6 is normal-highs, band 7 is ultra-highs. The purpose of the band linking is to avoid getting a lot of ultra-highs in this situation which wouldn't match the original.

Now this S might occur multiple times in a track, sometimes with lots of other instruments present (AGC is down a lot, S enters MB at a low level), and sometimes with almost no other sounds (AGC is up, S enters MB at a high level).

Very low input level
If the input level is VERY low, the sound won't change at all - output = input.

Low input level
If the input level is a bit louder, band 6 will be pushed down, but band 7 will stay at or close to 0.
Method 1 and 2 both will both cause band 7 to be pulled down with band 6 a bit, causing the ultra-highs to be reduced a bit.
Method 5 would cause band 7 to stay up completely, causing all the ultra-highs to stay.

Normal input level
Band 6 is pushed down a lot, band 7 is pushed down a bit without band coupling.
Method 1 and 2 both will both cause band 7 to be pulled down with band 6 a bit further, causing the ultra-highs to be reduced a bit.
Method 5 will cause band 7 to be pulled down as well, but probably a bit less than methods 1 and 2. The difference isn't that big though.

High input level
Method 1 and 2 don't care about volume levels, the relative amount of band 6 and 7 will always remain the same.
Method 5 will do the same thing it did at Normal input level. So this also gives the same result as at Normal input level.

So what's the difference between Method 1 and 2 then?
When the difference between band 6 and 7 gets bigger, the effect of the linking is reduced. Since in the example band 7 is a lot lower than band 6 anyway, this doesn't have to be a bad thing. Because the audio in band 7 will still be a lot softer than in band 6. We want to avoid the highs in band 7 to climb up to the same level as those in band 6, but we don't necessarily have to keep them down completely.


I think the following image shows the effect of the 3 different coupling methods on the attenuation of band 7 at different band 7 input levels

The leftest picture´s behavior is IMHO the best. But why has it to be in dB?
This behavior but with percentual coupling value would be perfect. So the effect gets smaller if th relative difference between bands is bigger.

That´s what I´ve tried to tell you...

linking would stop a band from going up more than a *percantage amount* above the level of the source band.

That´s what I meant.

So:
With 50% Linking, Band 6 has 12dB Audio:
Band 7 will not go above -6dB

Band 6 has 6dB Audio:
Band 7 will not go above -3dB

Furthermore the behavior of band 7 should be like in your leftest picture (Method 5).

You're saying two different things in your 2 posts above:
You're talking above relative difference here.
But this is not relative but based on absolute attenuation levels.


I don't like absolute levels because - see my description of the S tones above - it would behave completely different in the different scenario's, leading to big differences in sound depending on the MB input level.


Now, if I use relative difference levels as you wrote in your first reply, because the relative difference changes when the level of the band the we're linking to changes, there cannot be an angle as in image 1 (method 5), but it has to be gradual such as in image 2 or 3 (methods 2 and 1).

For me, a linking defined in dB would mean this:

With 6dB Linking, Band 6 has 12dB Audio:
Band 7 will not go above -6dB

Band 6 has 6dB Audio:
Band 7 will go to 0dB.

But that´s not good.

So I mean to define in percent, relative to the source band´s gain reduction.

I don't understand why in this case couldn't be such an angle like in the Method 5 picture. In this picture I assume a constant level of band 6, only band 7`s Level changes.
And when Band 6 stays at -12 (cause of the audio in it), then a linking value of 50% would be the same as 6dB - until Band 6`s gain reduction changes. If band 6 goes down to -6, the 50% linking value would be the same as 3dB.
And when the Level for this example stays constant, and only band 7 gets more audio from total silence in it until a high amount.