Smoothers and `smoothcc` #181

paulfd · 2020-04-13T14:10:05Z

This implements a linear and multiplicative smoother, trying to match a reasonable mix of behaviors between ARIA and consorts. The linear smoother uses a VA one-pole filter. The smoothing cutoff is a function of the sample rate, and it is quantized/tabulated between a minimum (a smoothing value of 1 with the default sample rate at 48 kHz) to a maximum (a smoothing value of 100 with a max sample rate of 192 kHz). We'd have to evaluate if this is enough.

Implement the smoothers with mapping to the smoothcc value and some tests.
Add a gain smoother on the voice for the overall amplitude (Implement dynamic crossfade #22); the smooth value is set at 10 for now.
Parse the bend_smooth opcodes and add a bend smoother for the pitch bend in the Voices.
On parsing in the Synth, check the max number CC parameters there are for each modifier target and adapt the number of smoothers in the Voices.
Jointly loop over the modifiers and the smoothers to compute the modifier envelopes in the Voices.

Closes #48, #22, #153

jpcima · 2020-04-16T19:04:58Z

Can you elaborate on what is the distinction between log and linear?
and why we need them? I lack the context to understand this.
("reasonable mix of behaviors between ARIA and consorts")

Also why's not simplest form of LPF not adequate, and why it's needed to tabulate the power function? (given smoothcc is supposed to be fixed?)

(sorry for lots of questions)

FrnchFrgg · 2020-04-16T19:26:50Z

As a remember when things are working, xfin_smoothccXX and xfout_smoothccXX would be great, even though they are not supported by LinuxSampler (which arguably is the most complete implementation on Linux to date), and maybe even not by ARIA/Sforzando.

paulfd · 2020-04-16T20:03:45Z

edited out

@FrnchFrgg this is linked to #22, and I added a gain smoother here which has this function more or less, but having smooth crossfades is one of the goal of this PR 🙂

jpcima · 2020-04-18T06:59:01Z

I don't find the filter to match to the RE'd specification.
Have you checked it?

jpcima · 2020-04-18T07:24:31Z

This formula matches it exactly.
It lacked a term 2 pi in the relation between the cutoff and time constant.

    smoothing = (smoothValue > 0);
    if (smoothing) {
        float tau = smoothValue * 3e-3f;
        float f = 1.0f / (twoPi<float>() * tau);
        float s = std::tan(pi<float>() * f / sampleRate);
        filter.setGain(s);
    }

paulfd · 2020-04-18T10:54:35Z

No I was planning to check it out now so thanks :)

paulfd · 2020-04-18T13:29:17Z

In the end it should reduce to this:

std::tan(1.0f / (2 * smoothValue * 3e-3 * sampleRate));

jpcima · 2020-04-18T15:28:57Z

Yes; in addition you can group all constants together.

std::tan((1.0f / (2 * 3e-3f)) / (smoothValue * sampleRate));

Do we need to tabulate this? I thought the smoothcc are set once and never changed.

paulfd · 2020-04-18T15:34:36Z

Well the filters belong to the voices, so the have to adapt to the currently played region for all smoothcc, and possibly sample rate. I guess we could set the smoothing filters upon loading the sfz file and make copies of the filter with their gain already set withing the voice, was this what you had in mind?

jpcima · 2020-04-18T15:42:18Z

If I understand this correctly..
it's supposed to be changed in case of:

loading sfz file
changing sample rate

None of these belong to DSP processing, hence why care to save a small fraction of processing time?

paulfd · 2020-04-18T16:16:06Z

No, it's changed when starting a voice, but I'll remove it and benchmark later if this creates problems.

jpcima · 2020-04-18T18:13:55Z

Feel free to cherry-pick 272b195, a test jack client

paulfd · 2020-04-18T23:45:22Z

You can use this to do further work on modulation if you wish.

jpcima · 2020-06-20T13:23:48Z

src/sfizz/Voice.cpp

+
+    const auto events = resources.midiState.getPitchEvents();
+    const auto bendLambda = [this](float bend) {
+        return region->getBendInCents(bend);


@paulfd, I'm trying to rebase this.
This function name is misleading. It returns a factor not cents.

Move modifiers and their helpers in their own files Add an enum class for modifiers and facilities to iterate over all possible targets Add smoothers to the voices and preallocate them Iterate over smoothers and modifiers jointly in the voices for each target

jpcima · 2020-06-20T15:24:54Z

I seem to have trouble with the following patch.
It's set to control panning left and right using a midi controller change.

When smoothcc is set and the note is keyed on, pan is going to start at the right and it will take a delay until the pan value establishes itself at center.

<control>
set_hdcc16=0.5

<region>
sample=*sine
ampeg_attack=0.02
ampeg_release=0.1
pan=-100
pan_oncc16=200
pan_smoothcc16=100

paulfd · 2020-06-20T15:29:02Z

I have no computer to try this until tomorrow, sorry. You mean that the value always start on the left on key presd even though the cc value is not changed? If so it's a problem, it means the panning value does not properly query the CC state at some point yeah. Jun 20, 2020 17:25:29 JP Cimalando <notifications@github.com>:

…

I seem to have trouble with the following patch. It's set to control panning left and right using a midi controller change. When smoothcc is set and the note is keyed on, pan is going to start at the right and it will take a delay until the pan value establishes itself at center. <control> set_hdcc16=0.5 <region> sample=*sine ampeg_attack=0.02 ampeg_release=0.1 pan=-100 pan_oncc16=200 pan_smoothcc16=100— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub[#181 (comment)], or unsubscribe[https://github.com/notifications/unsubscribe-auth/ADUFWQM2TCFRHVALIQNTTY3RXTIFDANCNFSM4MG7ML5Q]. [https://github.com/notifications/beacon/ADUFWQJ5UMNW42ACOI7JWPTRXTIFDA5CNFSM4MG7ML52YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOE2IJIVQ.gif]

jpcima · 2020-06-20T15:33:42Z

It appears the initialization of the filter which is incorrect. I 'll check it.

jpcima · 2020-06-20T16:07:15Z

The modulated unit has been changed to % after rebase, but the initialization had not followed.
This concerns amplitude, pan, width and position.
Fixed in e69365e

This was linked to issues Apr 13, 2020

Implement dynamic crossfade #22

Closed

Smooth CC formula #48

Closed

Add a small amount of smoothing on the amplitude/crossfade envelope #153

Closed

paulfd mentioned this pull request Apr 14, 2020

Solve some issues with DSmolken's drumkits #172

Merged

2 tasks

paulfd force-pushed the smoothers branch from 3295a79 to 2d989dd Compare April 14, 2020 22:26

paulfd force-pushed the smoothers branch 3 times, most recently from 7dae185 to 7dfb98b Compare April 18, 2020 23:25

paulfd marked this pull request as ready for review April 18, 2020 23:44

paulfd requested a review from jpcima April 18, 2020 23:44

paulfd added 4 commits June 20, 2020 14:35

Changes to envelope benchmarks

acba0ba

Add some missing functions to CCMap

114cd22

Add Defaults.h to Filterpool

832a5dc

Move readNote into the Opcode translation unit

cb3f534

jpcima reviewed Jun 20, 2020

View reviewed changes

paulfd added 2 commits June 20, 2020 15:27

Reduce the gain smoothing

cbb2341

jpcima and others added 8 commits June 20, 2020 15:27

Add the smoother demo client

fb19034

Trying to make tidy happy

0c68e40

Comments and licenses

b0bae1d

Remove the duplicate definition

e160e75

Make getBendInCents actually return cents

48de654

Fix some errors

377050a

Remove the old test file from CMake

0ccc57a

Fix the step/smooth CC ranges

997ae59

jpcima force-pushed the smoothers branch from b23aa25 to 997ae59 Compare June 20, 2020 13:58

Apply centsFactor at the other site where necessary

59c66b9

jpcima approved these changes Jun 20, 2020

View reviewed changes

jpcima added 2 commits June 20, 2020 18:00

Add missing includes

de9335b

Add percent normalization in initialization

e69365e

jpcima merged commit 02630ed into sfztools:develop Jun 20, 2020

paulfd mentioned this pull request Jul 5, 2020

Crossfade smoothers #313

Merged

paulfd deleted the smoothers branch January 21, 2022 10:24

Uh oh!

Smoothers and smoothcc #181

Smoothers and smoothcc #181

Uh oh!

Conversation

paulfd commented Apr 13, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jpcima commented Apr 16, 2020

Uh oh!

FrnchFrgg commented Apr 16, 2020

Uh oh!

paulfd commented Apr 16, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jpcima commented Apr 18, 2020

Uh oh!

jpcima commented Apr 18, 2020

Uh oh!

paulfd commented Apr 18, 2020

Uh oh!

paulfd commented Apr 18, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jpcima commented Apr 18, 2020

Uh oh!

paulfd commented Apr 18, 2020

Uh oh!

jpcima commented Apr 18, 2020

Uh oh!

paulfd commented Apr 18, 2020

Uh oh!

jpcima commented Apr 18, 2020

Uh oh!

paulfd commented Apr 18, 2020

Uh oh!

jpcima Jun 20, 2020

Choose a reason for hiding this comment

Uh oh!

jpcima commented Jun 20, 2020

Uh oh!

paulfd commented Jun 20, 2020 via email

Uh oh!

jpcima commented Jun 20, 2020

Uh oh!

jpcima commented Jun 20, 2020

Uh oh!

Uh oh!

Smoothers and `smoothcc` #181

Smoothers and `smoothcc` #181

paulfd commented Apr 13, 2020 •

edited

Loading

paulfd commented Apr 16, 2020 •

edited

Loading

paulfd commented Apr 18, 2020 •

edited

Loading