Improve Formant Filter User Interface

[fundamental]

Based upon feedback observed in a recent vlog by @unfa ( https://www.youtube.com/watch?v=Y93Jh1UIpys )

Notes:

• It's good to see that dynamic filter parameter updates are used for instruction
• There really should be some sort of preset phoneme formats setup. This might be a good option to throw in a random dropdown in the UI or it could be a good use for the currently underutilized and unintuivitive .xpz preset functionality in Zyn. When approaching this task see "Software For a Cascade/Parallel Formant Synthesizer" Dennis H. Klatt (1979) for a table of frequencies
• Displaying the live vowel position would be handy. Calculating out the live filter response would be too computationally heavy at a reasonable FPS (at least for my tastes). Showing the idealized center freq/gain/q should be enough (See Fig. 1,2). If live responses get implemented for formants they should be added for other filter types using the same "highlight the center freq" approach.
• ~13:00 Renaming and perhaps rescaling formant slowness to inertia would put zyn more in line with the conventions of other synths
• Formant filter parameters really should be updated live as well, and that could be a good task to direct any new contributors to
• There should be a visualization for the clamping and for the mapping of LFO/env to formant position. Originally I had a deadicated diagram for this, though I'd lean towards integrating this into Figure 3
• Perhaps add some 'auto-normalize' button to the interface
• 17:50 clipping issue, perhaps a non-linear element should be an option (both in formant and the 'analog' filter types as they're excessively sensitive to modulation). As a general note the UI should hint to the user in some way that state variable filters should perform better under heavy modulation (that said, I/someone should double check the interpolation method in the state variable filter as it might not be done correctly at the moment).
• 20:34 real units on the CF/octave sliders would be nice (though it is displayed elsewhere in the interface)
• 30:00->35:00 "Clunky to use", yep, you want to manipulate several points placed in R^3 which are in relationship to each other. For that I would propose the vowel/live view (Fig. 1,2) which will allow dragging formants around quickly in an abstract sense and Fig.3 for moving formants around in relation to other sequenced vowels. Additionally I would recommend adding the same grab support to the normal filter response graph. I originally had a set of notes for where grab points should be on filter response graphs and what their vertical/horizontal restrictions should be, but it looks like it was one of the features cut when time ran short on the UI dev sprint. These features should be revisited as it allows much faster and much more visual manipulation of parameters.

Figure 1. Display of vowel positions with hinting about band pass shape (non-numerically accurate). The user should be able to grab circles here and move with drag/drop operations and perhaps manipulate the alt parameter with scroll wheel.

Figure 2. Alternative Display of vowel positions

Figure 3. Display of vowel sequence. Emphasis put on sequence pos and frequency, alternative direction should show q or amplitude with the user selecting which one they want to see. The user should be able to grab circles here and move with drag/drop operations and perhaps manipulate the alt parameter with scroll wheel.

[unfa]

Wow, this looks like formant filter can be very intuitive and easy to use. This is far beyond what I could imagine as an improvement - a table of values would already be more readable, but this is even better.

I'm so glad my videos inspire improvements to ZynAddSubFX! Thank you for your amazing work, @fundamental !

[unfa]

Also:

The formant frequency response doesn't reproduce the notches that occur when using high-Q filters.

Example: https://youtu.be/Y93Jh1UIpys?t=23m58s

In my video you can see that Baudline shows a different spectrum than ZynAddSubFX, maybe it'd be good to correct for that too?

[fundamental]

Per the extra notches, those are an unintended side effect of the DSP implementation.

A full explanation would take a while, but essentially the formant filter is built by summing up the output of several band pass filters. Each filter produces one resonant peak, but summing two together does not give you (in the frequency domain) the sum of the magnitude responses. The phase information is also important and since different filters produce different (frequency dependent) phase responses, summing them results in other changes in the total frequency response. To fix this I think the simple solution might be a set of fractional delay filters to compensate for phase responses, but there might be a simpler solution that I'm overlooking as it has been a while since I've looked at the summed resonant band pass setup in a while. If such a solution is implemented, it will likely apply to SUBsynth as well.

[DerAndere1]

The phoneme-preset idea and the new visual drag/drop design looks great. Maybe the visualization in Fig. 1 is more intuitive than Fig. 2 because the formants are represented more like peaks in a spectrum. Actually the filter would be easier to understand if I it had only one vocal and relied on MIDI-learn/automation to modulate each of its parameters (cfreq, bandwith and amplitude of for each formant). But I understand that with 12 formants, this would end up in a huge list of parameters. So sticking with Zyn's concept (sweeping through a sequence of predefineable vocals) is probably better for handling.

Edit: At least in Fig. 3 The 1/q radius schould be more transparent and the handles in the center smaller. Otherwise the diagram might look crowded when 12 formants are active.

@unfa: Thx for all your helpful video tutorials! @fundamental: Zyn is a beast and it deserved your amazing new UI :)