1. I was thinking of a tool that can help a user with tuning multi-band plug-ins. The analyzer that presents a signal's dynamics in a dynamic-frequency plot. A simple (broad-band) version of it can be found in Wavelab7/Analysis/Loudness Distribution. I wonder, if it can be implemented in frequency domain. You could use a colour coding to present that data, like in a sonogram.
2. Could you add Max-Avg Difference type to your FFT analyzers? Maybe not ideal, but this graph can give us a good sense of signal's dynamics. For example, it could help with figuring out the band's distribution in multiband compressors. That is, if a part of a source's spectrum is more dynamic than the rest, then we can tune a special band for this region.
Have you tried e.g. AnSpec plugin with the "Density Mode" enabled in the Settings? Does it offer any insights? This display mode can be also used in Soniformer.
You can also enable the 2nd Spectrum MAX in SPAN which together with the AVG spectrum displays the dynamic range, at least in FFT estimation. But it does not look very clear, so you have to use spectrum smoothing.
Another inherent issue of such dynamics display is that higher frequencies usually have the most dynamics in any practical case. Lower frequencies have a far less average dynamic range. So, this skews the objective perception quite a bit.
Yes, I've tried all that.
One can easily cheat Density Mode by signal's timing. For example, a bass drum, at fixed velocity, playing 1/2 note will show higher dynamics than a hi-hat with a random velocity playing 1/16th notes. Thus, Density Mode is rather misleading for dynamics assessing.
A combination of MAX and AVG gives some insight. However, deciphering it is such a pain. It's like looking at FFT freq-cy analyzer without Slope control and trying to figure out a freq-cy response. If we could see the MAX-AVG curve, relative to the horizontal axis, it would be much easier.
That "inherent issue" means one simple thing - all those modes weren't created for dynamics evaluation but for freq-cy response.
But what is your solution to such Density Mode "cheating"? Because MAX+AVG spectrums has exactly the same issue - you can "cheat" it the same way.
I personally see an inherent, mathematical, problem of dynamic density evaluation.
It's MAX less AVG, not MAX plus AVG. I'm talking about the graph showing the difference between MAX and AVG. And here is why it can't be cheated by the same way as the Density Mode. Let's go back to the previous example.
Case 1: a bass drum and hi-hat play 1/2 notes at a fixed velocity (100). Then, MAX-AVG would be the same for the freq-cy regions corresponding to these drums. In turn, it would suggest no need for a multiband compression.
Case 2: a bass drum plays 1/2 notes with 100 velocity and a hi-hat 1/16 with the velocity between 100 and 50. In this case, MAX-AVG for the bass drum would be the same as in the case 1. However, MAX-AVG for the hi-hat would be higher because the AVG is lower now. You could even improve AVG calculating by excluding a silence out of equation (so it won't fall down in pauses).
As for Density Mode, my solution is - it must reflect only peak values until a user resets a meter. Here is how it should work, I think. Let's take a signal that starts from 0 and reaches 1, first. Next, it goes down to 0.3. Then, up to 0.5 and finally to 0 (If you draw it, it will look like a capital M). What will a user see on a meter? At the first part (0-1), the meter shows nothing. As soon as the signal enters the second stage (1-0.3), the meter must show a single bar at 1 and it must stay there. During the second (1-0.3) and third (0.3-0.5) stages, the user sees no changes on the meter, just that single bar at 1. As soon as the signal enters the 4th part (0.5-0), the meter must add one more bar at 0.5. Thus, the meter shows all peak values of a signal happening during some time (until the user resets it). You could also introduce some kind of bar colour brightness normalization. That is, coincident bars look brighter than single ones (that will make a "picture" similar to the one we can see now).
By looking at such a meter, we could judge a signal's dynamics (1-0.5 difference). Because we could easily figure out the max and min peak values. What we are interested, at setting multiband compressors, is a signal's dynamics on a freq-cy axis, not its maximum or average.
Basically, the mode, I described, reminds the Loudness Distribution meter in Wavelab. The difference is, Wavelab represents the loudness in time, but the Density Mode would give us an absolute number. From the user point of view, it's the same, I think, as both methods allow us to figure out a signal's dynamics.
But even if you implement it in AnSpec, for example, it will give us very poor picture because AnSpec filters are so broad. That's why I asked if it could be implemented in FFT. And yet, even AnSpec version is better than nothing!
One more thing. I don't insist on those methods (MAX-AVG, or Density Mode as I see it) as on a "true" dynamics metering. However, they can be a great help in tuning multiband compressors. After all, neither analog-like, nor FFT analyzers measure a true freq-cy response. And yet we use them every day with a great success.
I know that. And I know that the MAX mode is not perfect either. If you are after a mathematical precision, why do you use real time FFT analysis in your plug-ins, in the first place? I guess, it's because despite all their errors they still do a decent job. I'm not asking for a measuring tool after all, but for something that can give an idea of dynamics.
I'm simply trying to find a solution. I think, the best one is a loudness distribution in the freq-cy domain. Every mastering engineer would love to have it. But I can see you are not really keen. Never mind. I'll try to convince somebody else.