Waldorf Quantum/Iridium: keep the instant attack of a percussive sound

[douglas-carmichael]

This corrects an over-correction of my own, from #166.

The device cannot play an amplitude stage shorter than 0.07 s (y = 0.06 × 1000^value), so a shorter one collapses to an instant stage, which snaps the VCA open and clicks when the audio does not start at a zero crossing. #166 fixed that by lifting every short amplitude stage to 0.07 s.

Applied unconditionally, that lift is wrong for the normal case. A sampled instrument carries its attack in the audio and starts at silence, so there is nothing to click - the lift only fades the recorded attack in over 70 milliseconds and takes the strike off every piano, plucked string and drum.

An Akai S-1000 electric piano asks for 3 ms and was written as 70 ms, 23 times longer. It stopped sounding like a piano; that is how this was found.

An instant attack is normal on this device

The manual: "If Attack is 0 and Decay set to a medium value you get a percussive sound; if you turn up the attack, you get a soft-sounding start."

Measured over 141 presets written by an Iridium:

amplitude attack	presets
exactly 0 (instant)	50 (35 %)
above 0, below 0.07 s	35 (25 %)
0.07 s or longer	56 (40 %)

60 % use an attack shorter than the lift forced, and a third use exactly the instant stage the lift was written to avoid.

Change

The audio of the zones is measured, and the stage is lifted only when a zone really starts with a step which can be heard: more than 2 % of the peak level of that audio - the ratio at which the wrap of a loop is treated as audible elsewhere in the code (LoopClickDetector.MINIMUM_RELATIVE_STEP). Measuring against the peak of the same audio keeps it independent of the bit resolution.

The measurement runs only for a preset whose attack is short enough to be affected at all, so a preset with a genuine attack reads no audio. Unreadable audio keeps the safe, non-clicking variant. The release is unchanged: a note-off lands at an unpredictable point of the audio, so its step cannot be measured in advance.

Verified

117 presets of an Akai S-1000 CD-ROM converted to Quantum/Iridium:

	presets
instant attack restored	78
genuine longer attack, untouched	38
lift kept because the audio would click	1

The one that keeps the lift is TINESNPAD B, whose worst zone starts at 3.49 % of its peak (first frame 970). DIGIJAZZ, the electric piano, starts at 0.43 % and now gets its instant attack back.

Whole-disc conversion time is unchanged at ~4.8 s.

[git-moss]

@douglas-carmichael Please resolve the conflicts, I am unsure which code is the current one.

[douglas-carmichael]

Rebased onto current main - same situation, a textual collision rather than two versions of the same fix.

This branch predates #356, and both change the de-click of the amplitude envelope in WaldorfQpatCreator. #356 made the release always get the minimum, even when the source leaves it unset; this one makes the attack keep its instant rise when the material does not need de-clicking. Merged, both hold:

• release: declickAmpRelease (isAmplitude, ...) - unchanged from main, still applied even at zero
• attack: declickAmpTime (isAmplitude && !allowInstantAttack, ...) - only lifted when startsWithAudibleStep says the audio starts with a step

Current is main, and this adds the audio inspection on top. Verified on two sources with an identical 10 ms attack: one whose audio fades in from silence keeps AmpEnvAttack = 0.01 secs, one whose first sample is a large step is lifted to 0.07 secs. The renaming of the parameter of declickAmpTime from isAmplitude to declick is only because the flag is now more than "this is the amplitude envelope".