Step 01 / 5
Build a slow, detuned saw body #
Use ~soundStage = 1. The voice SynthDef makes three Saw oscillators at −7, 0 and +7 cents. Each has −13 dB gain and a slow ±2.2-cent sine drift. The outer voices drift at 0.226 Hz, while the center uses 0.17 Hz; their drift phases differ. Stereo weights of 0.87/0.5/0.13 put the outside voices toward opposite channels while sharing the center.
var amp = EnvGen.kr(Env([0, 1, -2.dbamp, 0], [0.85, 2.5, 2.4], \lin, 2), gate, doneAction: 2).pow(1.6);
The amplitude rises over 850 ms, decays toward −2 dB over 2.5 seconds, then holds until the gate closes. Release lasts 2.4 seconds. Raising the envelope to power 1.6 softens its middle levels and lowers its effective sustain to −3.2 dB. A short held note can release before either rise or decay finishes. Longer notes expose the slow beating.
Apply −8 dB instrument gain and a 0.65 multiplier before the shared effect bus. Velocity scales the result by 0.55 + 0.45 × velocity; it changes level without changing envelope times. One fixed 0.5 output multiplier applies to every stage.
Sound after step 1 — WebM/Opus
Next: Add moving pulses and serial filtering →Step 02 / 5
Add moving pulses and serial filtering #
Use ~soundStage = 2. Add one Pulse to each saw at −16 dB, tuned another +5 cents above that voice. A single cirrusMotion SynthDef supplies the shared 0.09 Hz sine motion. Each note reads its control bus so overlapping notes agree on pulse-width movement.
var width = 0.29 + (0.025 * In.kr(motionBus));
var cutoff = 850 * (2 ** ((0.8 * filterEnv) + (0.15 * (midi - 60) / 12)));
var filtered = LPF.ar(BPF.ar(stereo, cutoff, 1 / 0.85), 6200);
The pulse width travels between 0.265 and 0.315. filterEnv rises over 1.8 seconds, falls toward 0.6 over 3 seconds and releases over 2.4 seconds. At MIDI 60, the bandpass moves from 850 Hz to about 1480 Hz at the envelope peak. Its rq is bandwidth divided by center frequency, so 1 / 0.85 gives a broad band rather than a narrow whistle. Key tracking adds 0.15 octaves of cutoff per octave of played pitch. The following 6200 Hz lowpass softens the upper harmonics.
Sound after step 2 — WebM/Opus
Next: Thicken the stereo image with ensemble →Step 03 / 5
Thicken the stereo image with ensemble #
Use ~soundStage = 3. In cirrusEffects, highpass the ensemble input at 220 Hz. Feed three cubic-interpolated DelayC taps from the left channel, the mean of both channels and the right channel.
var times = 0.020 + (SinOsc.kr(0.24, [0, 2pi/3, 4pi/3]) * 0.0036);
var taps = DelayC.ar([hp[0], hp.sum * 0.5, hp[1]], 0.04, times);
Each tap moves between 16.4 and 23.6 ms at 0.24 Hz. The three phases are separated by one-third of a cycle. Blend 78% dry with 22% ensemble; each wet side uses 75% of its outside tap and 25% of the center tap. Then multiply the side component by 1.05. Increasing delay movement makes the pitch bending stronger; increasing the wet mix makes the moving image more prominent. This shared processor keeps overlapping voices in the same ensemble field.
Sound after step 3 — WebM/Opus
Next: Add a diffuse stereo room →Step 04 / 5
Add a diffuse stereo room #
Use ~soundStage = 4. Delay the room input by 28 ms and send it to the two-input FreeVerb2. Request its wet output, then blend that output with the unpredelayed ensemble.
var pre = DelayN.ar(signal, 0.028, 0.028);
var room = FreeVerb2.ar(pre[0], pre[1], 1, 0.84, 0.55);
var reverberated = (signal * 0.75) + (room * 0.25);
Room size is 0.84 and damping is 0.55. These dimensionless controls shape this algorithm's diffuse tail; room size 0.84 is not equivalent to Csound feedback 0.84, nor does damping 0.55 specify a 3300 Hz cutoff. The 25% wet blend retains the pad's moving center. More damping darkens the circulating high frequencies; a larger room value lengthens the lingering texture. The room remains active after the individual note Synths finish their releases.
Sound after step 4 — WebM/Opus
Next: Center the lows and finish the output →Step 05 / 5
Center the lows and finish the output #
Use ~soundStage = 5. Convert the reverberated stereo signal to mid and side, highpass only the side at 180 Hz, then reconstruct left and right. This reduces low-frequency spread while preserving motion above the cutoff.
var finalMid = mixed.sum * 0.5;
var finalSide = HPF.ar((mixed[0] - mixed[1]) * 0.5, 180);
var focused = LeakDC.ar([finalMid + finalSide, finalMid - finalSide]) * 10.9.dbamp;
LeakDC removes DC offset and the +10.9 dB gain restores level after the filtered effect chain. The additional fixed 0.5 output multiplier remains in place. LeakDC differs from the longer-window Csound dcblock2, particularly at very low frequencies. The result is a soft, slowly moving pad with a stable low center and a diffuse upper image.
Signal order matters: the motion source runs first, the voice group writes to a private stereo bus, and the effects Synth reads that bus afterward. The full script maintains that order so the same instrument can play overlapping notes.
Sound after step 5 — WebM/Opus
Audio & settings ↓